Search
Author
Title
Vol.
Issue
Year
1st Page

Journal of Animal Science - Animal Production

Technical note: Evaluation of a system for monitoring individual feeding behavior and activity in beef cattle

 

This article in JAS

  1. Vol. 93 No. 8, p. 4110-4114
     
    Received: Jan 21, 2015
    Accepted: May 15, 2015
    Published: July 2, 2015


    1 Corresponding author(s): eftimsit@ucalgary.ca
 View
 Download
 Share

doi:10.2527/jas.2015-8947
  1. B. Wolfger*,
  2. A.V. Mang*,
  3. N. Cook,
  4. K. Orsel* and
  5. E. Timsit 1*
  1. * Department of Production Animal Health, Faculty of Veterinary Medicine, University of Calgary, Calgary, AB, Canada T2N 4N1
     Agriculture and Agri-Food Canada, Lacombe Research Centre, Lacombe, AB, Canada T4L 1W1

Abstract

Behavioral observations are important to detect illness in beef cattle. However, traditional observation techniques are time and labor intensive and may be subjective. The objective was to validate a system for monitoring individual feeding behavior and activity in beef cattle (Fedometer [FEDO]; ENGS, Rosh Pina, Israel). Sixteen steers (initial BW ± SD = 326 ± 46 kg) were fitted with data loggers (FEDO) on their left front leg and housed in a pen with a feedbunk equipped with an antenna emitting an electromagnetic field that reached 30 ± 2 cm in front of the feedbunk. Feedbunk attendance (duration of visit and frequency of meals) measured by FEDO was compared with live observations (27 observational periods lasting between 72 and 240 min; mean 126 min). Lying time and frequency of lying bouts were compared with previously validated accelerometers fitted to the hind leg (10 steers equipped for 10 to 12 d; HOBO Pendant G Acceleration Data Logger [HOBO]; Onset Computer Corporation, Pocasset, MA). Step counts were compared with video recordings (15 observations for 6-min intervals in 6 steers). Concordance correlation coefficients (CCC), accounting for repeated measures, and limits of agreement were computed. Comparison between FEDO and observed time at the feedbunk yielded a CCC of 0.98 (95% confidence interval [CI] 0.97–0.99). All 68 meal events observed were recorded by FEDO. However, FEDO recorded 4 meal events during the 27 observational periods that were not observed. Lying time measured by HOBO and FEDO were highly correlated (CCC = 0.98; 95% CI 0.97–0.99). However, frequency of lying bouts measured by FEDO was only moderately correlated to HOBO (CCC = 0.71; 95% CI 0.63–0.77); FEDO underestimating the number of lying bouts (on average, 0.4 fewer bouts per 6 h). Step count by FEDO was moderately correlated to video observations (CCC = 0.75; 95% CI 0.49–0.89); FEDO overestimating the number of steps (on average, 5 more steps per 6 min). In conclusion, the FEDO system accurately measured duration of feedbunk attendance, frequency of meals, and lying time. However, it overestimated the number of steps and underestimated the frequency of lying bouts.



INTRODUCTION

Feeding behavior and activity are frequently monitored to evaluate health status of beef cattle; animals are considered ill when they are off feed or have decreased activity (Weary et al., 2009). Traditionally, such monitoring has been solely based on visual observation, although it is time and labor intensive and may be subjective (Weary et al., 2009).

Various automated systems for monitoring cattle feeding behavior and activity have been developed (Theurer et al., 2013). These systems continuously collect data and eliminate potential observer bias (Weary et al., 2009). For example, feedbunk units of the GrowSafe system (GrowSafe Systems Ltd, Airdrie, AB, Canada) continuously measure individual feeding frequency, duration, and intake (Wolfger et al., 2015). Additionally, various accelerometers can accurately measure activity (i.e., standing and walking) and lying behavior (Robert et al., 2009; Mattachini et al., 2013).

Recently, a system measuring both cattle individual feeding behavior (i.e., bunk visit frequency and duration) and activity (i.e., lying time, number of lying bouts, and numbers of steps) has been marketed (Fedometer [FEDO] system; ENGS, Rosh Pina, Israel). This system uses a data logger attached to the front leg that, in addition to monitoring activity, detects the presence of cattle at the feedbunk. Combining both behaviors in 1 system may provide a more accurate assessment of the true health status (Theurer et al., 2013) and be more cost efficient. However, evaluation of the accuracy of the system has not been reported.

The objectives of this study were to determine accuracy of FEDO system to measure 1) feedbunk attendance (duration of attendance and frequency of meals) compared with live observations, 2) lying time and frequency of lying bouts compared with a previously validated accelerometer, and 3) step counts compared with video recordings.


MATERIALS AND METHODS

This study was conducted at the Agriculture and Agri-Food Canada, Lacombe Research Centre (Lacombe, AB, Canada) during August 2013. All procedures were reviewed and approved by the University of Calgary Animal Care Committee (AC13-0104).

Sixteen Hereford × Angus steers (initial BW ± SD = 326 ± 46 kg) from the Lacombe Research Centre beef herd were used. Steers were housed in an outdoor dirt floor pen (60 by 60 m) topped with wood shavings. They had ad libitum access to fresh water and were slick-bunk fed with 100% barley silage once daily at 0900 h. Although 95% of feed was eaten by 2200 h, little, low-quality feed was available until the next feed delivery.

At the beginning of the study, the outer wall of the 55 m feedbunk was equipped with an active antenna (ENGS) that emitted an electromagnetic field every 8 s that extended 30 ± 2 cm in front of the feedbunk (Fig. 1). Each steer also had a FEDO data logger strapped to the distal lateral aspect of their left metacarpus. This data logger had a rigid plastic housing of 68.8 by 50.7 by 26.5 mm and weighed 75 g. It measured g-force in the x-, y-, and z-axes and also passively detected the electromagnetic field emitted by the antenna installed at the feedbunk. Every 6 min, the data logger actively and wirelessly transmitted 1) binary data (presence at the feedbunk [yes/no], lying [yes/no], and change from standing to lying [yes/no]) and 2) counts of steps and electromagnetic field detection to a receiver that was connected to an on-farm computer via an electric cable. Proprietary software (ENGS) converted 1) g-force readings into lying time, number of lying bouts, and number of steps and 2) counts of electromagnetic field detection into duration of feedbunk attendance and meal frequency. Duration of feedbunk attendance was calculated based on the actual count of electromagnetic field detected by the data logger, whereas meal frequency was calculated based on a 5-min rule (Schwartzkopf-Genswein et al., 2002) with feedbunk visits less than 5 min apart summarized into 1 meal.

Figure 1.
Figure 1.

Fedometer (FEDO; ENGS, Rosh Pina, Israel) logger recognizes electromagnetic field transmitted by antenna.

 

To evaluate the accuracy of the FEDO system for measuring duration of feedbunk attendance and frequency of meals, 2 trained observers recorded feedbunk attendance of steers during 27 observation periods lasting between 72 and 240 min (mean 126 min; Table 1). Observation periods were performed over 11 d between 0430 and 2200 h. During each period, 2 observers randomly selected 2 steers and recorded the start and finish of feedbunk visits, defined as a steer having its head on top of the feedbunk with the neck between the chest and neck rail (Mendes et al., 2011). To enable comparison with the output of the proprietary software, observed duration of feedbunk attendance did not include any nonfeeding periods and frequency of feedbunk visits was reported as meals (i.e., feedbunk visits less than 5 min apart summarized into 1 meal). Interobserver variability was tested before the study and repeated 5 times during the study. Both observers independently recorded total duration of feedbunk attendance on the same 2 steers (n = 12 observational periods; r = 0.99). The distance of the electromagnetic field emitted by the antenna (30 ± 2 cm) was verified before the study and weekly during the study period using a “tester tag” provided by the manufacturer. This “tester tag” was held 15 to 20 cm off the ground (simulating the fetlock joint) and flashed when the electromagnetic field was detected.


View Full Table | Close Full ViewTable 1.

Observational times per steer used for comparison with Fedometer (FEDO; ENGS, Rosh Pina, Israel)

 
Steer no. Observations, h Feeding, min No. of meals1
1 14 236 10
2 6 107 5
3 2 44 4
4 2 11 1
5 6 148 7
6 8 36 3
7 4 46 2
8 11 98 5
9 9 111 5
10 4 15 1
11 7 62 3
12 7 114 3
13 12 165 4
14 6 66 2
15 17 286 6
16 14 154 7
1Meal frequency was calculated based on a 5-min rule (Schwartzkopf-Genswein et al., 2002) with feedbunk visits less than 5 min apart summarized into meals.

To determine if lying time and lying bout frequency measured by the FEDO system were accurate, accelerometers previously validated for measuring lying behavior in cattle (HOBO Pendant G Acceleration Data Logger [HOBO]; Onset Computer Corporation, Pocasset, MA; Ito et al., 2009; Bonk et al., 2013) were installed on 10 steers. These devices were programmed to record g-force on the x-, y-, and z-axes at 1-min intervals and were attached to the left hind leg above the fetlock, as described by Ito et al. (2009). The logger’s memory of 64 kB had the capacity to record 21,800 3-dimensional data points (Moreau et al., 2009). After 10 (5 steers) or 12 d (5 steers) of data collection, HOBO were removed and stored data were downloaded. Onset HOBOware software (Onset Computer Corporation, Bourne, MA) was used to convert g-force readings into degrees of tilt. Macros in Microsoft Excel (Excel version 14.4.1; Microsoft Corporation, Redmond, WA) translated degrees of tilt into lying time and number of lying bouts.

To evaluate the accuracy of step counting by the FEDO system, the pen was equipped with a video surveillance system consisting of 8 infrared day/night varifocal cameras (SONY Color CCD; Sony Corporation, Tokyo, Japan) and a recording computer. During 6-min video recordings (n = 90), 1 observer counted steps (forward or backward movement on left front leg) on 6 randomly selected steers.

For data analysis, time at the feedbunk and number of steps were summarized by observation periods. Lying time and number of lying bouts were summarized by 6-h periods to account for potential diurnal differences. Number of steps was log transformed to achieve normal distribution. The R (R Foundation for Statistical Computing, Vienna, Austria) package CCCRM was used to calculate concordance correlation coefficients (CCC) adjusted for repeated longitudinal observations on lying time, number of lying bouts, number of steps, and feeding time (Carrasco et al., 2013). With this statistical package, steer was included as a random effect. Therefore, the outcome refers to the median CCC observed among all steers. All other statistics were performed using Stata version 11.2 (StataCorp, College Station, TX). Residuals of paired measures were represented in Bland and Altman graphs (Bland and Altman, 2010).


RESULTS

During the 27 observation periods that were compared with FEDO records, 68 meals were observed (Table 1). Twenty-one observations of individual steers did not have any feeding events. Additionally, eight 2-h observations were lost due to technical failure of the FEDO system. On average, observers recorded 1.4 (SD 0.6) meals during each observation period. In comparison, the FEDO system recorded 72 meals with an average of 1.5 (SD 0.6) meals per steer per observation period. The FEDO system detected all 68 observed meals (100% sensitivity) but also detected 4 additional meals not classified as feedbunk attendance by observation (i.e., neck was not between the rails; 94% specificity). Observed time at the feedbunk was, on average, 1.1 min shorter than feedbunk time recorded by FEDO (Fig. 2a). Concordance correlation between FEDO and observed time at the feedbunk was 0.98 (95% confidence interval [CI] 0.97–0.99).

Figure 2.
Figure 2.

Difference between Fedometer (FEDO; ENGS, Rosh Pina, Israel) and a) feeding behavior recorded with live observations, b) lying time during 6-h periods recorded by validated accelerometers (HOBO Pendant G Acceleration Data Logger [HOBO]; Onset Computer Corporation, Pocasset, MA ), and c) number of steps counted with video observations in 6-min observations.

 

Data for HOBO and FEDO were available for 10 steers and a total of 259 complete 6-h periods. Due to a power outage, 7.5 d of FEDO records were lost. Additionally, 1 HOBO detached after 1 d. In total, 1 steer was used twice for HOBO measurements, that is, 51 6-h periods and between 25 and 28 6-h periods were available for comparison for all other steers. The average lying time per 6 h recorded by HOBO was 171 min (SD 114 min) with 2.5 (SD 1.7) lying bouts, whereas FEDO recorded an average of 172 min (SD 114 min) of lying with 2.1 (SD 1.7) lying bouts per 6 h (Fig. 2b). The CCC between the 2 accelerometers was 0.98 (95% CI 0.97–0.99) for lying time and 0.71 (95% CI 0.63–0.77) for lying bouts.

Based on video observations, steers took, on average, 11 steps per 6 min (range 0–189), whereas the FEDO system recorded 16 steps (range 0–362; Fig. 2c). Concordance correlation between observed and FEDO recorded number of steps was 0.75 (95% CI 0.49–0.88).


DISCUSSION

Compared with the dairy industry, few monitoring systems have been validated in beef cattle. Notwithstanding, automated real-time monitoring of beef cattle behavior is crucial for farm expansion, especially with increasing labor costs (Wagner et al., 2014). In the present study, we evaluated a new system (FEDO) that can measure both feeding behavior and activity of beef cattle. This system accurately measured duration of feedbunk attendance, meal frequency, and lying time. However, it overestimated the number of steps and underestimated the frequency of lying bouts.

Although the FEDO system detected proximity to the feedbunk and not active feeding, this system accurately measured feeding behavior. The measured time spent at the feedbunk was highly correlated to observations and only 4 additional meals were monitored by FEDO. This high accuracy was consistent with a report that most feedbunk visits (i.e., 89%) by feedlot steers were associated with active feeding (Schwartzkopf-Genswein et al., 1999). Although the accuracy of meal frequency measurement could be further improved by using head-locking systems (Bach et al., 2004), the FEDO system allowed all cattle to feed simultaneously, which reflects natural cattle behavior (Rook and Huckle, 1995).

The FEDO system accurately measured the lying time but underestimated the frequency of lying bouts. This underestimation can be explained by the long sampling interval for lying behavior (i.e., 6 min). Indeed, sampling intervals >2 min have previously been described as inadequate to predict the number of lying bouts (Mattachini et al., 2013). Undeniably, with a long sampling interval, some short lying bouts will not be recorded. Accurate measurement of lying bout frequency can be important. For example, there were more lying bouts in castrated calves given pain medication (flunixin meglumine) compared with castrated calves without pain medication (Mintline et al., 2014).

The moderate correlation between step counts measured by FEDO and video observation was not expected. Indeed, other validated accelerometers (Icetag; IceRobotics, Edinburgh, UK) provided sensitivities ranging from only 0.26 to 0.29, although it provided high specificity (Mattachini et al., 2013). This was attributed to the higher sampling frequency reported by the FEDO manufacturer, that is, 1,000 Hz versus the 8 Hz reported for the Icetag (Mattachini et al., 2013). Therefore, the FEDO system can be useful to detect changes in the number of steps, which has been associated with estrus (Firk et al., 2002), lameness (Alsaaod et al., 2012), or respiratory disease (Hanzlicek et al., 2010).

In addition to simultaneously monitoring 2 behaviors (i.e., feeding and activity), this system transmits data wirelessly, enabling real-time monitoring of behaviors and overcoming the limited storage capacity of previously validated systems. For example, in the present study, the HOBO loggers had to be removed and replaced after 12 d due to limited storage capacity (associated with a 1-min transmission interval). However, it was noteworthy that we lost 7.5 d of data due to inadequate storage on the on-farm computer, which could have been avoided with data storage in the data logger or a server-based automated data storage.

In conclusion, the FEDO system accurately measured duration of feedbunk attendance, meal frequency, and lying time. However, it overestimated the number of steps and underestimated the frequency of lying bouts. Therefore, it could be used in research settings and field applications to accurately monitor feeding and activity in beef cattle.

 

References

Footnotes


Comments
Be the first to comment.



Please log in to post a comment.
*Society members, certified professionals, and authors are permitted to comment.