Validation of a Radio frequency identification 1 system for tracking location of laying hens in 2 a quasi-commercial aviary system

Cage-free housing

subset of animals per flock was tracked. We applied an RFID tracking system to monitor all 23 1125 laying hens of a flock, which were divided into 5 pens of 225 birds each in a barn with 24 an aviary system. In each pen, 26 antennas were placed on the edges of three tiers and in the 25 litter. For validation purposes, 3 hens in 2 connected pens were fitted with colored backpacks. 26 They were recorded on video and their location throughout the pen was taken from the video 27 and compared with registrations from the RFID system. For 93% of compared transitions, the 28 RFID data matched the observational data regarding the tier or litter whereas the value fell 29 to 39% for specific antennae. When the antennae on the litter were excluded for the 30 validation, the match on tier-level was at least 98% but on antenna-level it remained lower 31 than 50%. The sensitivity of the detection of tiers/litter but not antennae differed among the 32 three hens. We conclude that the RFID tracking system was suitable for studying the 33 movement pattern of individual hens among tiers in an aviary system in a reliable way but 34 tracking birds on the litter needs to be improved.

Introduction 42
Cage-free housing systems for laying hens may contain tens of thousands of animals. Although 43 considered welfare-friendly, cage-free housing systems including aviaries are known to entail risks 44 concerning health (e.g. parasites, infections) and animal welfare (e.g. damaging behaviours like feather-45 pecking and cannibalism) ( Ali, et al., 2020). In principle, aviaries are designed to offer essential functional areas to the hens like aerial 47 perches for (nighttime) roosting, secluded nest areas for laying, and a litter area for exploratory behavior 48 and dust-bathing. However, individual birds access these areas to a different extent (Rufener, et al., 2018)

49
which is known to correlate to various health risks (Rufener, et al., 2019;Ali, et al., 2020). 50 Tracking individuals in large groups of identically looking laying hens is a challenge that can either be 51 attempted by visually marking the animals or by an electronic tracking system (for reviews see Li  nestbox, lower) as well as in the litter, three antennae were put side-by-side joining at the short end.

77
Additional antennae were placed on each side of the wintergarden although not evaluated in this effort.

78
As a test trial for future experiments on a large number of birds, all birds in 5 pens of a barn with 20 pens 79 with 225 birds per pen were fitted with a glass tag (HITAGS 4x22mm, 125KHz, HTS256) in a custom-80 developed leg band (Fig. 2). If a tag was detected by an antenna a time stamp and the identities of tag and 81 antenna were written into a .csv file every 0.1 s. However, if a tag remained on the same antenna for a 10 82 s period, the registration was not repeated in order to limit the size of the generated files. The maximal 83 vertical reading distance of all antennae was about 15 cm and the horizontal reading distance was close to 84 0 cm. Three hens in a pen that was connected at the level of the litter to a neighboring pen for free 85 movement between the two pens wore color-coded back-packs that were visible on video recordings. One 86 observer watched videos recorded between April 21 st and 29 th , 2021 until a total of 10 hours of video were 87 scored, on which at least one hen with a custom-made backpack (Fig. 3) for identification was visible. Based 88 on the recorded video (30 fps), the location of those hens walking, standing, or sitting on the antennae and 89 the pen at each change of location with the respective video time stamps was entered into a spreadsheet. 90 Additionally, the observer noted whether the identification of the hen was certain or uncertain due to poor 91 visibility. of the aviary, and in the correct pen. In particular, several variables were extracted for the closest RFID 108 registrations in time that matched the hen (see Table 1), and tests were performed to compare the RFID 109 data and coded observations (see Table 2). 110 The results were entered in a confusion matrix to calculate the sensitivity of the RFID system (true

Closest antenna
The antenna code recorded in the closest RFID event as stated above.

Closest tier
The tier corresponding to the closest RFID event as stated above.
Closest side The side of the aviary corresponding to closest RFID event as stated above.

Closest pen
The pen recorded in the closest RFID event as stated above.  Variable Meaning

Side correction needed
The side as recorded by the observer had to be corrected by a second person because there was an obvious error in coding by the observer (antenna coding for the wrong side was used, based on the antenna and side mismatch). This did not involve RFID data.

Closest antenna matches
The antenna code recorded in the RFID event for the observed hen with closest timestamp to the coded video observation time is the same as observed.
Same antenna within 1 min.
Same as above but the RFID data matches the observed antenna within a 1 min. window. It is not necessarily the match closest in time.
Closest tier + side + pen match The tier recorded by the observer matches the tier (and side) of the antenna code of the RFID registration closest in time for the respective bird.
Same tier + side + pen within 1 min.
Same as above but the observer matches the tier (and side) within a 1 min. cutoff. It is not necessarily the match closest in time.
Closest pen matches The pen recorded by the observer matches the pen of the antenna code of the RFID registration closest in time for the respective bird.
Closest side of the aviary matches The side of the aviary recorded by the observer matches the side of the aviary of the antenna code of the RFID registration closest in time for the respective bird.

127
From the video files, 304 locations of the three birds were detected of which the observer was certain 128 (75.6% of all sightings of birds on antennae). Of these, in 91 % of the cases, the correct tier, side, and pen 129 of the aviary was detected within 1 min. by the RFID system (Tab. 3a). In all but 7 cases, this was also the 130 closest RFID detection in time. In 1 case, the correct tier, but at the opposite side of the aviary was indicated 131 by the RFID system. The correct tier regardless of the side of the aviary and the pen was detected in 93% 132 of the cases. Sensitivity fell precipitously to 39% when the focus was detecting the correct antenna within 133 one minute. In 3% of the cases a wrong pen was indicated and in 2% the wrong side of the aviary. 134 When the registrations of the litter were excluded, detection was much better (Table 3b). All 135 sensitivities on tier-level were between 98 and 99% whereas the sensitivities regarding the correct antenna 136 within tier remained below 50%. 137 The registration of the RFID system was on average 1.6 s (Stderror = 1.9 s) earlier than the video time 138 stamp if the tier identified by the RFID and observer matched and 3.6 s.

141
Of the three hens, each accounted for 43.4% (132), 34.9 (106), and 21.7% (66) of all registrations. The 142 hens differed in the sensitivity of the registrations relative to tiers including the litter but not antennae 143 when all tiers included the litter were analyzed. However, with the exclusion of the antennae on the litter, 144 hens only differend when the same antenna within 1 min. was considered. The difference was due to the 145 two birds with the fewer registrations. Of those, one hen had about 5 times more correct than incorrect 146 registrations of the antenna within 1 min. and the other bird had twice as many incorrect than correct 147 registrations of the antenna within 1 min. 148 149

158
The detection rate of birds on the different tiers and in the litter of an aviary system was very high and 159 comparable to other efforts using different RFID systems in poultry with either equal or greater sensitivities 160 ( than the reliability of 99% of the active low-frequency tracking system by Montalcini et al.( 2022). Although 165 overall sensitivity was high, the correct antenna was detected in less than 50% of the cases. The poor 166 detection can be explained by the fact that the antennae were positioned adjacent to each other so that a 167 tag likely could be read intermittingly by both antennae when the hen sat on both. The problem of birds in 168 between antennae has also been a problem for other efforts (van der Sluis et al., 2020). In addition to this 169 problem 'within' pens, the problem could also persist 'across' pens. As pens were adjacent, antennae of 170 one pen also touched antennae of the neighbouring pen leading to registrations in the 'wrong' pen. In case 171 that pens are connected and the movement of birds between pens is studied, this likely error would need 172 to be addressed. For instance, to resolve the problem of false pen registrations, the edges of antennae at 173 the extreme sides of the pen can be physically blocked (Ringgenberg, et al., 2015). In either case, our efforts 174 suggest the benefits of such a validation to help improve accuracy and determine potential solutions. More 175 critically, our results also indicate that the present set-up did not yield adequate precision to tell where 176 across the 225 cm wide tier the hen was located, i.e. we achieved only the registration of the tier and side 177 with acceptable levels. Given our validation results, tracking individuals at the side/tier level is possible, 178 but a higher resolution may be necessary depending on the research question. 179 In 20 instances, the RFID registration did not match the correct tier. In all but 1 of these cases, the bird 180 was seen on the litter but the antenna immediately above the litter on the first tier was recorded instead. 181 In one mismatch, the hen was seen on the antenna on the highest tier and it was recorded on that tier but 182 on the other side of the aviary. In each of these cases, the hen likely moved faster than the registration 183 window, e.g. up to the first tier / down to the litter or underneath the aviary to the opposite side. Speed 184 of registration has been shown to be a problem with fast moving laying hens with a similar RFID system 185 (Gebhardt-Henrich, et al., 2014). For the current validation, an improved system with faster registration 186 was used. However, it is possible that very fast moving hens may still be missed. The resolution of the 187 timestamp in the csv file generated by the RFID system was 0.1 s. Since it is impossible to synchronize the 188 video system with the RFID system with this accuracy, the time difference between the RFID registrations 189 and the video time stamps are not surprising. 190 Tier-specific, incorrect registrations also likely result from the set-up of the aviary and the spatial 191 configuration of the antennae. Interestingly, we found almost no mistakes in terms of tier recordings 192 except in the litter. The decreased sensitivity of the litter is likely because birds can more easily enter the 193 area without coming into contact with an antenna. In contrast, a bird transitioning between the upper and 194 nest box tiers would have to step onto an antennae at the edge of each zone. As a solution to improve 195 sensitivity in the litter, we have doubled the number of antennas there with a later setup.

197
The three hens differed in the sensitivity of the registrations of the tier and the positions where they 198 were observed in the aviary. One hen was mostly seen on the litter while another on the uppermost tier. 199 The hen with the lowest sensitivity scores had fewer registrations but was seen both on the litter and the 200 uppermost tier. The sample size of three hens is too low to draw any conclusions whether certain 201 individuals would differ in the sensitivity of the registration of tiers. However, it is feasible that such a 202 difference exists due to variations in an individual's behavior (e.g., flying or jumping across antennas) or 203 preference of certain locations in the aviary which are less reliably registered on the antennas. In our 204 dataset the difference in the sensitivities likely resulted from differences in litter use because sensitivities 205 on tier-level no longer differed among hens when registrations of antennae on the litter were excluded. 206 Differences in the registration of the antennae within one min. were due to the 2 hens with fewer 207 registrations and the cause is unknown.

209
This validation was done before the start of the full experiment so we do not have tracking data from 210 other hens for this period. However, density of hens and equipment of the pen was the same as in the 211 following studies except the addition of a second row of antennas on the litter in following experiments.

213
It is important to note that a gold standard to determine the positions of the hens does not exist. The 214 observations from the videos were error prone and in almost one quarter of observations the combination 215 of antennas and side of the aviary were impossible and had to be corrected. Mistakes while coding videos 216 occur like in other easy tasks that do not require a high level of conscious attention esp. when the observer 217 is disrupted (Morrison, 2021). In addition, the antennae on the tiers of the aviary could be clearly seen on 218 the videos whereas the exact positions of antennae on the litter were less obvious because they were 219 covered by litter. This could have added to the lower sensitivities of detection on these antennas. 220 Furthermore, it was difficult to synchronize our video and RFID systems with the resolution of less than 1 221 s. because both systems were not connected to the same network. 222 In conclusion, the employed RFID system reliably detected the position of hens on the different tiers in 223 an aviary in a reliable way but tracking birds on the litter needs to be improved. 224

Acknowledgements 225
We thank Masha Marincek to observe the hens on the videos. Abdelsatar Abdul Rahman installed and 226 serviced the RFID system daily. Numerous helpers were involved in catching and banding birds. 227  Project fund. We also thank Hendrix Genetics BV for the in-kind contribution of laying hens. 235