Water consumption and wastage behaviour in pigs: implications for antimicrobial administration and stewardship

Animals, accommodation and facilities

The study was conducted in a naturally ventilated grower/finisher building with concrete floors, side curtains, and a central ridge vent and passage way, accommodating 1 184 female pigs and 2 335 immunologically castrated male pigs in 90 pens. Fifteen immunologically castrated male pigs weighing approximately 88 kg (based on a BW-for-age chart) were arbitrarily selected and placed in a study pen measuring 3.3 m wide by 5.98 m long (19.7 m²). Approximately 39% of the pen’s total floor area was solid concrete and 61% was concrete slats. The pen was equipped with two nipple drinkers positioned one metre apart along the east wall. Each nipple drinker was oriented horizontally, parallel to the fence, and had a stainless steel bowl beneath it to capture water spilled by the pigs. Each drinker had partitions extending 41 cm from the fence into the pen. One of these partitions was a solid material; the other was a metal mesh structure. The pen was also equipped with a Nedap* Prosense single-space automatic feeding station occupying 0.92 m² of the pen area, located in the north-east corner of the pen (Fig. 1).

1. Download : Download high-res image (145KB)
2. Download : Download full-size image

Fig. 1. Layout of study pen accommodating the group of 15 pigs, showing the dimensions and positions of the drinkers and feeder*. *Nedap Livestock Management, Parallelweg 2, 7141 DC Groenlo, The Netherlands.

The water source provided to pigs in the building was surface water, pumped from a river and stored in a farm dam. A water sample was collected aseptically from the first drinker during the study for chemical and microbiological analysis at a commercial environmental testing laboratory and found to be well within acceptable standards for all chemical parameters tested (Edwards and Crabb, 2021) (Water quality analysis report is provided in Supplementary Table S2). Pigs were fed ad libitum with a pelleted ration, formulated to meet the nutritional requirements of finisher pigs as specified by the National Research Council (2012).

Water and feed metering systems

A turbine flowmeter with a Hall effect pulse generator (Vision 1005 2F66, 1/4″ connection, Badger Meter Europa GmbH, Neuffen, Germany) was fitted to the water line supplying the two drinkers. The flowmeter had a measurement range of 0.1–2.5 L/min, a K-factor of 22 000 pulses/L, and an accuracy of ±3%. A tipping bucket water gauge with a 40 ml measuring bucket and a dual output reed switch (HyQuest Solutions TB6/40) was positioned on the external side of the east wall of the pen, directly behind each of the two drinkers. A pipe from the bottom of the bowl below the nipple drinker directed spilled water above the tipping bucket (Fig. 2). The water gauges were designed to measure a maximum water flow rate of 3 L/min, well in excess of the flow rate from each drinker (1.25 L/min). Measurement of water wasted in 40 ml increments was adequate for the purpose of calculating water wastage on an hourly timescale. The tipping bucket water gauges were used in preference to a turbine flowmeter, as water spilled by pigs at drinkers was likely to be contaminated with feed particles from their mouths and saliva, which could interfere with the function of the rotor in a turbine flowmeter.

1. Download : Download high-res image (193KB)
2. Download : Download full-size image

Fig. 2. Water metering system used in the study pen to measure the water consumption and wastage behaviour of the group of 15 pigs.

The system recorded a timestamp with water in-flow in pulses/min from the turbine flowmeter and out-flow in the number of tips/min from each tipping bucket water gauge. These data were stored locally and periodically uploaded to a remote data server as a.csv file for processing. In-flow (ml/min) was calculated by dividing the number of pulses/min recorded by the turbine flowmeter by 22 000. Out-flow (ml/min) at each tipping bucket was calculated by multiplying the number of tips/min by 40. The single-space automatic feeding station contained an RFID reader, feed trough and weighing platform. Each time a pig entered the feeder, it was identified by its RFID ear tag, and its BW, feeding duration and feed intake were recorded. Data were transmitted in real time to Nedap’s proprietary data management system, Nedap Velos. After an adaption period of 14 days, the water input and output from the two drinkers and feed consumption from the automatic feeding station were recorded over a period of 27 days. At the completion of the study, the mean BW of the pigs was approximately 109 kg.

Data aggregation and analysis

Water used for each hour of each day was calculated as the sum of water used for each minute of that hour. Data on water wasted for each of the two tipping bucket water gauges were combined, and total volumes tipped for the pen were calculated for each time period. Where tips occurred in consecutive minutes, total volume tipped in the pen was calculated for these minutes combined. For each time period with tips, total water usage since the previous tip was calculated and the proportion of that water that was wasted was calculated. Water wasted in each minute of the day was then calculated, assuming the proportion wasted was constant since the previous tip. Water wasted for each hour of each day was then calculated as the sum of water wasted for each minute of that hour. Water consumed in each hour of each day was then calculated as water used less water wasted in that hour. All feed dispensed by the feeding station was assumed to have been consumed by pigs. Feed consumed for the pen group was recorded for each time period when feed was dispensed, with the feeding period start time (hour and minute of day) and duration of feeding in seconds recorded. Feed used in each hour of the day was estimated from these data. The rate of feed consumption (g/sec) was assumed to be constant throughout each feeding period. Temperature and humidity data, and times of sunrise and sunset on each day of the study, were obtained from a weather station 7 km from the farm (Australian Government Bureau of Meteorology, 2021).

Data on the daily water use and wastage of pigs and their association with feed consumption were subjected to statistical analysis in R and visualised using ggplot2 in R. Models for pigs’ water use and wastage patterns over each 24-h period were estimated using the software package brms (Bürkner, 2017), which provides an interface to fit Bayesian generalised (non-) linear multivariate multilevel models in R using the probabilistic programming language Stan (Chatzilena et al., 2019). We opted to use a Bayesian inference method, as used in some previous studies of pig water use patterns (Madsen and Kristensen, 2005, Jensen et al., 2017, Dominiak et al., 2019a), rather than a frequentist method, as it offered several advantages. These included a hierarchical structure that offered great flexibility, with the ability to readily use datasets of varying sizes and to specify and analyse complex hierarchical models, and a more coherent expression of uncertainty. As it was a hierarchical generalised additive model (HGAM) (Pedersen et al., 2019), brms assumed that the water usage and wastage of pigs in a given hour were dependent on their water usage and wastage in previous hours, and identified changes in patterns of water used and wasted within each 24-h period over 27 successive days as pigs gained weight. The HGAM enabled the relationships between the explanatory variable (hour of day) and the response variables (water used and wasted) to be described as smooth curves (“smoothers”). For water usage per kg pig BW per hour, we generated a single common smoother plus group-level smoothers (each of the 27 days of the study) for all observations. Each group-specific smoother had its own smoothing parameter and therefore its own wiggliness (Pedersen et al., 2019). The same approach was also used for water wastage per kg pig BW per hour.

The model is characterised by the general equation:

is the day-level intercept, representing inter-day variability. Observations of unobservable low value were defined as left-censored, with an arbitrarily nominated left-censoring limit of 1 millilitre per pig per hour. This was implemented to ensure that observations of unobservable low value were retained in the model (Canales et al., 2018). The effective sample sizes were evaluated and increased as necessary and the ‘adapt_delta’ argument was altered to ensure that divergent transitions did not occur in the results. For each model run with brms, for each smooth term, group and population-level effects, chain convergence was assessed with the Rhat statistic and a value of 1.00 was achieved, indicating that the chains had converged to a common distribution (Vehtari et al., 2021). The final version of the code used to fit the models with brms in R is provided in Supplementary Material S1.

In-water antimicrobial dosing events (days 8 and 9)

Periodic, metaphylactic in-water dosing events were routinely conducted in the building. Two of these events coincided with days 8 and 9 of the study. All pigs (including those in the study pen) were administered an antimicrobial, lincomycin, by the farm manager according to the consulting veterinarian’s prescription, applying dosing practices routinely used on the farm. On each day, dosing commenced at approximately 0700 and ceased at 1500 (8 h duration). A value for water wastage of 45% was factored into the dosing calculations performed prior to the dosing event, as was done routinely for all grower/finisher buildings on the farm fitted with nipple drinkers. To achieve the prescribed target dose of 8 mg/kg BW of active lincomycin, an antimicrobial dose of 14.55 mg/kg BW was therefore administered, assuming that 6.55 mg/kg BW (45%) would be wasted.

All components of the water metering system installed in the pen for this study functioned reliably for the study period, without any interruption to data flow. Particulates in the dam water supplying the pig building were prevented from disrupting the rotor within the turbine flowmeter by two screen filters installed in the water line, before the water holding tank and between the water holding tank and the turbine flowmeter (Fig. 2). The model of turbine flowmeter selected for the study proved to be capable of measuring short flows at very low flow rates so that all volumes of water drawn by pigs when interacting with the two drinkers in the pen were recorded. The tipping bucket water gauges remained in position under the spout of the pipe draining water from the bowl beneath each drinker, ensuring that nearly all volumes of water wasted were recorded. Feed particles and saliva present in water collected by the tipping bucket flow gauges did not disrupt their function.

Daily pig water usage and wastage and association with daily feed consumption

Water usage per pig per day increased slightly over the study period, as was expected as the pigs gained weight. The water used per kg BW per day was steady over the study period (median, 84.7 ml; interquartile range, 10.9; range, 49.4–123 ml/kg) (Fig. 3a). Water wasted per kg BW per day was also steady (median, 29.3 ml; interquartile range, 5.3; range, 15–41.1 ml/kg). The proportion of water used by pigs daily that was wasted (median, 36.5%) was moderate compared to those reported in previous studies (Supplementary Table S1) and quite stable from day-to-day (interquartile range, 3.5%) (Fig. 3b).

1. Download : Download high-res image (607KB)
2. Download : Download full-size image

Fig. 3. (a) Water wasted and consumed by the pen group of 15 pigs per day of the 27-day study period (ml/kg BW); (b) Proportions of water used by the pen group of 15 pigs per day of the 27-day study period that were wasted and consumed. *Feeder malfunction on days 11–13. ^{^}Feeder malfunction on days 20–21.

The proportion of the total daily water usage of pigs that occurred during daylight hours (i.e. between sunrise and sunset) was greater (median, 72.8%; interquartile range, 7.2%; range, 66.2–91.3%) than that of feed consumption (median, 60.7%; interquartile range, 8%; range, 35.5–67.6%) (Fig. 4a). While feed consumption per pig per day trended upward slightly over the study period as the pigs gained weight, median feed consumption per kilogram BW was more steady (median, 24.8 g; interquartile range, 5.9; range, 14.5–31.5 g/kg). The water used:feed ratio (median, 3.5; interquartile range, 0.5; range, 2.8–6.4) and water consumed:feed ratio (median, 2.3; interquartile range, 0.4; range, 1.8–4.1) were also similar to those reported in a previous study (Bigelow and Houpt, 1988) and were very consistent from day-to-day (Fig. 4b).

1. Download : Download high-res image (287KB)
2. Download : Download full-size image

Fig. 4. (a) Proportions of water used and feed consumed by the pen group of 15 pigs per day over the 27-day study period during daylight hours^#; (b) Water:feed ratios, based on water wasted, water consumed and total water used (sum of water consumed and water wasted) by the pen group of 15 pigs per day over the 27-day study period (ml/g). Days 11–13 and 20–21 were excluded from analysis as the feeder malfunctioned on these days, so accurate feed consumption data were not available. ^#based on sunrise and sunset times for Corowa airport, Australian Government Bureau of Meteorology (7 km from farm).

Aberrations in water use by the pigs were seen in association with malfunction of the automatic feeder on days 11–13 and 20–21 of the study (Fig. 3a). At 0648 on day 11, the feeder became totally inoperative for a period of 51 h 6 min until it was repaired and returned to normal operation on day 13. During this period, pigs were fed manually 3 kg of the pelleted ration per pig on the floor at 0730 each day. On days 11 and 12, a reduction in the volume of water used by the pigs was observed. On day 13, following manual feeding of pigs at 0730, the feeder returned to normal operation at 0954. Pigs’ feed consumption on day 13 was therefore greater than normal and corresponded with the highest daily water usage measured during the study. Restriction in the quantity of feed available to pigs occurred again on day 20 due to malfunction of the automatic feeder. As was seen on days 11 and 12, a reduction in the volume of water used by the pigs was observed on day 20. The next day (day 21), following manual feeding of pigs at 0730, the feeder returned to normal operation at 0916. Pigs’ feed consumption on day 21 was therefore greater than normal, and corresponded with the second-highest recorded water usage during the study.

Patterns of water used and wasted over each 24-h period

Smoothers generated in brms for pigs’ water use and wastage observations for the 27 days by time of day showed that both had uni-modal patterns, peaking in the mid-late afternoon at approximately 1400–1700 (Fig. 5). Over the 27-day study period, peaks for water used and water wasted of pigs tended to shift 1–2 hours earlier in the afternoon. During the mid-late afternoon, when water use was higher, a slightly greater proportion of the water used was wasted. Overall, the model fit was sound (R²: 0.67; 2.5th credible limit: 0.60; 97.5th credible limit: 0.72). While the model did not perform well on days 13 and 21, the two days of aberrantly high water use during the study, this did not impact on the R² and 2.5th and 97.5th credible limits of the model.

1. Download : Download high-res image (978KB)
2. Download : Download full-size image

Fig. 5. Smoothers showing estimated water usage and water wastage by the pen group of 15 pigs (ml/kg BW per hour) within each day over the 27-day study period, in chronological order. In each smoother, the band edges indicate the limits of a 95% credible interval. Observed data for water used and water wasted in each hour of each day are represented by dots.

In-water antimicrobial dosing events (days 8 and 9)

The median water wastage rate measured in the study pen over the 8-hour dosing period (0700–1500) on days 8 and 9 was 38%, which was 7% less than the value of 45% factored into the dosing calculations by the farm veterinarian and farm manager prior to the in-water dosing event conducted in the building. Pigs would therefore have consumed 9 mg/kg of the antimicrobial instead of the target dose of 8 mg/kg, equating to 1.125 of the target dose, an over-dose of 12.5% (Fig. 6).

1. Download : Download high-res image (429KB)
2. Download : Download full-size image

Fig. 6. Proportion of target dose of antimicrobial consumed by the pen group of 15 pigs at alternative actual water wastage rates during a dosing event, based on the value for water wastage (% of water used) factored into the dosing calculations prior to the event. The value for water wastage factored into the dosing calculations by the farm veterinarian and farm manager prior to the in-water dosing event conducted in the building was 45%. The circle indicates the median water wastage rate of 38% that was measured in the study pen over the 8-hour dosing period (0700–1500) on days 8 and 9 (38%).

Patterns of water used and wasted over each 24-h period

Water use patterns within each 24-h period of cohorts of pigs reported in published studies have varied widely in the timing and the amplitude of the peak(s) (Turner et al., 2000, Madsen and Kristensen, 2005, Brumm, 2006, Villagra et al., 2007, Kashiha et al., 2013, Rousseliere et al., 2016, Chimainski et al., 2019, Muniz et al., 2021, Misra et al., 2021). This may be for three reasons. First, the behavioural patterns of a particular cohort of pigs reared in a particular building, including its drinking pattern, may be influenced by intrinsic factors such as the pigs’ gender, genetics, BW and health status, and many extrinsic factors such as the building type and design, pen group size and stocking density, drinker type, the number of drinkers and feeders in each pen and their spatial arrangement, water flow rates from drinkers, water quality, diet specifications, day-length, building temperature, humidity and air quality and flow (Villagra et al., 2007). Second, the method used to collect water use data varied between studies (equipment used, frequency of data measurements). Third, the statistical approach (frequentist or Bayesian) and specific method used to estimate each cohort of pigs’ water use pattern varied between studies.

In agreement with other researchers (Madsen and Kristensen, 2005, Dominiak et al., 2019a, Dominiak et al., 2019b), we found that aggregating data on water used and wasted per minute into hourly data points was a good compromise between losing excessive data points and having so many data points that the computational load was too great for a mid-specification personal computer to perform Bayesian inference through simulation. The use of group-level smoothers (each of the 27 days of the study) with the common smoother enabled the model to capture variation between days in the patterns of water used and wasted.

The two occasions during the study when the automatic feeding station malfunctioned highlighted the close association that exists between pigs’ water use and feed consumption. The close alignment of the uni-modal patterns of total water use and wasted water was expected. In other cohorts, we have observed shifts in peak water use of up to 3 hours over their occupancy in a building (unpublished studies). The study was conducted under cool to mild conditions in late winter (median minimum daily ambient temperature of 4 °C, median maximum daily ambient temperature of 16 °C). If the study had been conducted under warm to hot summer conditions (i.e. above the thermal comfort zone of finisher pigs of 10–25 °C), the proportion of the water used that was wasted may have been substantially higher in the afternoon and early evening than at other times, if pigs spent increased time using drinkers to wet their skin as an evaporative heat loss strategy. This warrants further investigation. If an in-water antimicrobial dosing event was conducted under such circumstances, many pigs would be under-dosed unless a higher, accurate value for wastage (