Acoustic analysis: A novel way to measure livestock grazing behavior

Final Report for GS14-130

Project Type: Graduate Student
Funds awarded in 2014: $10,981.00
Projected End Date: 12/31/2015
Grant Recipient: Virginia Tech
Region: Southern
State: Virginia
Graduate Student:
Major Professor:
Gabriel Pent
Dept. of Crop and Soil Environmental Science, Virginia Tech
Major Professor:
John Fike
school of Plant and Environmental Sciences, Va Tech
Expand All

Project Information


Wideband acoustic recordings were verified as an accurate method for documenting sheep foraging behavior, and time lapse cameras were deployed in place of GPS loggers to confirm sheep behavior due to logger inaccuracy. Preliminary results indicate that lambs in open pastures spent less time lying down in comparison to lambs in the silvopastures. Contrary to expectations, lambs in open pastures spent more time grazing with higher bite rates, though these grazing events were confined to mornings and evenings. Lambs in honeylocust silvopastures had highest afternoon body temperatures. Despite these variabilities, lambs in the honeylocust silvopastures had the highest weight gains.


This material is based upon work that is supported by the National Institute of Food and Agriculture, U. S. Department of Agriculture, under award number GS14-130.


Project Relevance to Sustainable Agriculture


Human population growth is increasing global demands for food and fiber production at the same time that the agricultural landbase and the resources for agricultural production are in decline (1-3). Concern about climate change and dwindling natural resources has led to a public push for decreasing environmental contamination (4).

Agroforestry offers opportunity to fulfill the demands both for increasing food and fiber production, while also improving environmental quality through increased nutrient utilization, reduced sediment runoff and erosion, and increased carbon sequestration (5-9). These diversified systems are managed so as to take advantage of positive interactions among system components and thus create greater output than monocultural production systems. Simple monocultures of pasture and animal systems have appealed to livestock producers in the past, as this simplifies management. However, the sustainability of such reductionist methods is being called into question (10) and systems that take advantage of complexity and diversity will be more productive and sustainable.

Silvopasture, one of five agroforestry practices, integrates trees with pasture-based livestock systems and provides both short- and long-term returns from the same land base (11). Livestock in silvopasture systems can benefit from shade in summer and from shelter from wind in winter. The trees in turn benefit from the managed livestock presence through amplified nutrient cycling and weed suppression (11). Along with potential benefits to food and fiber production, silvopasture systems may have increased soil organic matter with improvements in microbial health and nutrient cycling (12), greater water storage (13), and improved nutrient capture and retention (14). These factors, coupled with improved soil conservation and nutrient utilization, result in regional watershed benefits (15). Silvopastures sequester more carbon than timber plantations or pastures, an advantage with global ramifications (16). Along with these production and environmental benefits, silvopasture systems may require less nutrient and herbicide inputs; increase and diversify marketable productions; and produce aesthetically-pleasing landscapes that add value to farms and rural economies.

This project is part of an on-going effort to develop sustainable silvopasture production systems. My research addresses the animal behavior and welfare dynamics of silvopasture production and will develop tools to enable researchers to implement and monitor these systems. This project also addresses the productivity of these systems, a component critical in convincing farmers of the economic sustainability of silvopasture. As part of an existing SARE grant, we will monitor system production metrics (forage yield and nutritive value and animal performance). Support for this graduate student grant will allow us to meet the following objectives and better understand the mechanisms supporting animal performance in silvopastures.


Statement of Problem, Rationale, and Justification


The purpose of this project is to compare behavior and well-being of grazing livestock in deciduous silvopastures with that from open pasture systems. Our previous SARE-supported research has shown that tree species have differential effects on pasture composition; forage yield and nutritive value responses can be positive or negative depending on tree age, stand density, and slope position, among other factors (17, 18). This has not translated into differences in animal performance among systems, however.

Few studies have explored how temperate silvopastures designed and managed with deciduous trees affect both the forage base and the performance and behavior of grazing ruminants. Forage production and nutritive value vary quite widely depending on tree species and management, making relationships to animal performance more difficult to determine. Thus, while deciduous silvopastures may differ from open pastures in terms of forage yield, composition, or nutritive value, these responses do not necessarily track differences in animal performance between silvopastures and open systems (19-21).

Recent research (22, 23) with lambs grazing in walnut- and honeylocust-based silvopasture systems suggests animal performance is comparable to that from open pastures, even when forage yield is reduced. However, the mechanisms behind these responses have not been clearly defined. Some data suggest that increased forage nutritive value compensates for lower forage mass in silvopastures (e.g. 21), but lower soluble carbohydrates (18) and only moderate, variable responses in terms of fiber digestibility (23) challenge this idea. Altered animal behaviors – such as grazing time, rumination, standing, and lying – and consequences to energy expenditure may thus be more important drivers of the similar animal gains observed between open and silvopasture systems.

Animal behavior in silvopasture has not been well studied. Heat load may change activities and intensify stresses experienced by animals in open pastures, thus increasing time and energy spent in behaviors to stabilize body temperature. Ambient temperatures are lower and less variable in silvopasture systems; thus, animals may experience more time with conditions suitable for grazing and increase dry matter intake (DMI) (24). Distinguishing between reduced energy needed for maintenance vs. greater opportunity for grazing in pasture systems is challenged by the limited tools for monitoring grazing animals, and current methods involve time-intensive observations.

As part of a larger, SARE-supported study evaluating forage production and animal performance in silvopastures, this project will generate detailed information on animal behavior and heat load in silvopastures and open pasture systems. Specifically, we will: 


1) Use novel acoustic monitoring systems to determine lamb grazing behavior; 

2) Apply GPS tracking systems to determine shade distribution patterns and animal mobility and position relative to shade in silvopastures; 

3) Monitor lamb temperatures with temperature sensors; and 

4) Integrate information on forage quantity and quality with spatial and temporal information on grazing behavior and body temperatures to understand the effects of silvopasture dynamics on animal performance.


This research will increase the body of information regarding factors affecting animal performance in silvopastures and support the development of these sustainable land management systems.


Literature Cited


  1. Tilman, et al. Proceedings of the National Academy of Sciences, 108(50), 2011: 20260-20264.
  2. Pimentel, et al. Science, 194(4261), 1976: 149-155.
  3. Tilman, et al. Nature, 418(6898), 2002: 671-677.
  4. Chen, et al. Proceedings of the National Academy of Sciences, 108(16), 2011: 6399-6404.
  5. Nair. Netherlands: Springer, 1998.
  6. Wallace. Agriculture, ecosystems & environment, 82(1), 2000: 105-119.
  7. Tscharntke, et al. Biological Conservation, 151(1), 2012: 52-59.
  8. Young. Outlook on Agriculture, 19(3), 1990: 155-160.
  9. Montagnini & Nair. Agroforestry systems, 61(1-3), 2004: 281-295.
  10. Lyson. Trends in Biotechnology, 20(5), 2002: 193-196.
  11. Sharrow, et al. In North American Agroforestry: An Integrated Science and Practice, by Garret, 105-131. Madison: American Society of Agronomy, Inc., 2009.
  12. Chander, et al. Biology and fertility of soils, 27(2), 1998: 168-172.
  13. Sharrow. Agroforestry systems, 71(3), 2007: 215-223.
  14. Michel, et al. Plant and soil, 297(1-2), 2007: 267-276.
  15. Shrestha & Alavalapati. Ecological Economics, 49(3), 2004: 349-359.
  16. Sharrow & Ismail. Agroforestry Systems,60(2), 2004: 123-130.
  17. Buergler, et al. Agronomy journal, 97(4), 2005: 1141-1147.
  18. Buergler, et al. Agronomy Journal 98(5), 2006: 1265-1273.
  19. Peri, et al. Proceedings of the New Zealand Grassland Association 63, 2001: 139-147.
  20. Lehmkuhler, et al. Agroforestry systems, 59(1), 2003: 35-42.
  21. Kallenbach, et al. Agroforestry systems, 66(1), 2006: 43-53.
  22. Fannon-Osborne. M.S. Thesis. Virginia Polytechnic Institute and State University, 2012.
  23. Fannon-Osborne, et al. In preparation, 2014.
  24. Mitloehner & Laube. Journal of Animal and Veterinary Advances 2(12), 2003: 654-659.


Project Objectives:

  1. Develop and apply novel acoustic monitoring systems to quantify lamb grazing behavior in terms of prehensive biting events and rumination;
  2. Quantify lamb body temperatures diurnally; 
  3. Determine diurnal behavior (and shade utilization in silvopastures) using GPS tracking systems and remote sensing technology; and 
  4. Integrate information on forage quantity and quality with spatial and temporal information on grazing behavior and body temperatures to understand the effects of silvopasture systems on animal performance.



Materials and methods:

The 12-week grazing study was performed at the Whitethorne Agroforestry Demonstration Center at Virginia Tech's Kentland Farm (Blacksburg, Virginia). Trees for the silvopasture treatments were planted in 1995 and thinned to a final density in 2012 (about 25 stems per acre). Black walnut and honey-locust trees were used to create two silvopasture systems treatments for comparison with open pastures. Each system is 0.67 acres and replicated three times in a randomized complete block design.

Pastures and silvopastures were divided into four paddocks and were further subdivided for rotational grazing management. The honeylocust silvopasture and open pasture systems were stocked with 7 lambs while the black walnut silvopasture was stocked with 6 lambs, which were blocked by weight and sex at the beginning of the study. Lambs were weighed with a portable scale for two consecutive days at initiation and conclusion of the study and once each period. Once the individual animals had been assigned to an experimental unit they remained in that unit for the remainder of the study. Fecal samples from three lambs within each experimental unit were taken at each weigh date to assess for Haemonchus contortus infection rates. FAMACHA scores from every lamb were taken every two weeks after the first period and any lamb scoring a 4 or higher after the first period or a 3 and higher after the second period was dewormed with levamisole.

Forage mass was estimated using a rising plate meter before and after every rotation interval; a regression curve was made by double-sampling at every other measurement interval. Samples for nutritive value estimates were collected by hand plucking forage at every other measurement interval. Nutritive value measures (neutral and acid detergent fiber, crude protein, in vitro total dry matter digestibility, and total nonstructural carbohydrate levels) will be determined by standard laboratory procedures. Composition of the vegetation was determined every period using a modified, six-class Daubenmeyer approach. 

Three sheep in each experimental unit within a block were fitted with a Roland R-05 recorder (Roland Corporation, Los Angeles) and a Sennheiser ME-2 lavalier microphone (Sennheiser Electronic Corporation, Germany). The recorder was housed in a water-resistant bag attached to a body-harness. The lavalier microphone ran from the recorder and along a halter, to the front of the halter near the mouth of the sheep. The devices remained on the nine animals for one day and then were transferred to sheep in another block the following week for three consecutive weeks of measures in each period. The device recorded WAV files at a 16 bit resolution and 48 kHz sampling rate. Typical recording events lasted from 6:30 to 21:30 before the recorders were removed from the sheep, except in the chance of substantial rainfall when they were removed before the precipitation occurred. During the second and third periods, each week before sampling the equipment was fitted on the lambs to be sampled and the lambs were videotaped to provide synchronized audio and video files of prehension events. Acoustic files from one sampling date were reduced to monaural and a high-pass filter (600 Hz cutoff frequency, 48 dB/octave rolloff) was applied to remove low frequency noise using the Audacity software ( The prepared files were then analyzed in the program, SIGNAL and GRASS (Engineering Design, Berkeley), to quantify bite events. The software parameters were calibrated by manual determination of bite events from the synchronized video and audio files. Detection threshold was determined manually by listening to each files to find a series of bite events and using the parameter visualization function of GRASS to determine an adequate threshold to distinguish high frequency prehension events from low frequency noise still present in the files. GRASS reported start and end times, voltage, and squared voltage for each event meeting the manually determined parameters.

Ewe vaginal temperatures were measured by implanting a DST micro-T temperature logger (Star-Oddi, Iceland) into a blank controlled internal drug release device (Zoetis, Florham Park) and sealing the device with electrical tape. A communications box (Star-Oddi, Iceland) was used to download the data from the device to a computer for analysis by a software from the same company. The three sheep in each experimental unit fitted with the acoustic devices were also equipped with the vaginal temperature loggers for three days in a given sampling week before being transferred to sheep in another block the following week. Data from the devices were downloaded and the devices were sterilized and refit for deployment each week before sampling. Each recorder was calibrated the previous year to determine the correction factor necessary to precisely report temperatures to a set reference device. These correction factors were applied to each data point and the data was compiled and averaged by treatment and sampling period.

Due the low accuracy of the GPS loggers (3 m) used the preceding year, Moultrie D-500 trail cameras (EBSCO Industries, Inc., Birmingham) were programmed to take time-lapse photos every 60 seconds and were set on a tree, fence post, or PVC pipe to visually encompass the entire subpaddock containing the sheep being sampled during a given week. Before sampling, the sheep of interest were marked with pink, orange, or blue spray paint to distinguish the sheep in the time-lapse photos. The devices collected photos the same days that the acoustic data was being collected. The photos were processed sequentially by recording the behavior (standing up, lying down, grazing, drinking water, eating mineral) and shade utilization (in the shade, in direct sunlight, overcast) of each sheep at each minute.

PROC MIXED in SAS (SAS Institute, Inc., Cary), using a repeated measures model with unstructured variance, was used to conduct an ANOVA of treatment effect on average daily gains (ADG) of the lambs and total animal gain of each experimental unit.


Research results and discussion:

ADG was highest for the lambs in the honeylocust silvopastures (least squares means = 0.1418 kg/day; standard error = 0.01116 kg/day), followed by the ADG of lambs in the black walnut silvopastures (least squares means = 0.1229 kg/day; standard error = 0.01222 kg/day), and by the ADG of lambs in the open pastures (least squares means = 0.1040 kg/day; standard error = 0.01127 kg/day). The ADG of lambs in the honeylocust silvopastures was significantly higher than the ADG of lambs in the open pastures (P=0.0206), though the ADG of lambs in the black walnut silvopastures was not significantly different from the ADG of lambs in the open pastures (P=0.2592) and the ADG of lambs in the honeylocust silvopastures (P=0.2596). There was a significant treatment by period interaction effect on ADG (P=0.0191; Figure 1), most likely a consequence of high H. contortus loads (Figure 2).

 Due to the lower stocking rate of the black walnut silvopastures, it is more informative to evaluate animal gains by system rather than by individual animal performance (Figure 2). Gain per experimental unit remained highest for the lambs in the honeylocust silvopastures (least squares means = 27.8 kg/period; standard error = 2.12 kg/period), followed by the gain per experimental unit of the open pastures (least squares means = 20.8 kg/period; standard error = 2.20 kg/period), and finally by the gain per experimental unit of the black walnut silvopastures (least squares means = 18.7 kg/period; standard error = 2.22 kg/period). The gain per experimental unit of the honeylocust silvopastures was significantly higher than the gain per experimental unit of the open pastures (P=0.0254) and the gain per experimental unit of the black walnut silvopastures (P=0.0046), though the gain per experimental unit of the black walnut silvopastures was not significantly different from the gain per experimental unit of the open pastures (P=0.5173).

 Though FAMACHA scores were taken every other week beginning after the first period to determine which lambs should be treated with levamisole, lamb anemia, presumably due to H. contortus infection, remained an issue throughout the study (Figure 2). This was a likely cause for depressed sheep weight gains (Figure 1). For example, after the second period when lambs in the open pasture had the highest H. contortus fecal egg counts, the ADG of lambs in the open pasture system was 0 kg/day. ADG for the open pasture lambs increased after the third period, corresponding with a decrease in H. contortus fecal egg counts. Statistical analyses of correlation should provide some clarity on whether high H. contortus infection rates was responsible for overall low weight gains.

Pre- and post-graze calibration models for the prediction of forage mass by the rising plate meter were likely influenced by patchy trampling effects by the sheep under high forage availability scenarios, resulting in higher intercepts for the honeylocust and open pasture post-graze models (Figure 4, Table 1).

Until proper correction factors are determined, post-graze estimates will likely be inflated. Thus, pre-graze forage mass data are presented alone (Figure 5; Table 2). Throughout the season, forage availability was consistently lowest in the black walnut silvopastures. As a result, the stocking rate of the black walnut silvopastures was kept one animal lower (~25-30 kg) per experimental unit. There was little difference in forage availability in the honeylocust silvopastures and open pastures.

Results from the vaginal temperature sensors demonstrate as expected that the ewes in the open pastures have the largest diurnal change in temperatures, with large temperature gains occurring between early morning and late evening (Figure 6). Body temperatures of the open pasture ewes also tracked more closely with the temperature humidity index (THI), presumably becuase they are more exposed to ambient conditions without tree modulation effects. However, it was not expected that the ewes in the honeylocust silvopastures would have the highest peak body temperature in July and August. Statistical analyses remain to be completed through nonlinear modeling of the data to determine the significance of these trends. Interaction between treatment and month, or period, also remains to be investigated. Whether these higher variations in temperature for the ewes in the open pasture systems correlates to higher energy expenditure or lower animal comfort remains to be seen.

 Due to the amount of time required to document ewe behavior from the trail camera photos by minute, only one sampling date (August 4, 2015) was analyzed for this report. From these preliminary results, ewes in the black walnut silvopasture appeared to spend the most time in the shade compared to ewes in the other systems (Figures 7). Ewes in the open pasture spent more time standing up than lying down compared to the ewes in the other systems as they were presumably trying to dissipate heat more efficiently by maintaining higher proportions of body surface area exposed to wind. This phenomenon indicates a higher level of stress and energy expenditure for the open pasture ewes. These trends are also evident when examining total time that each ewe spent in each activity (Figure 8). It is important to note that late in the afternoon as shade from bordering trees covered the open pasture experimental unit, the ewes in the open pasture were able to access to shade around 16:30 (Figure 7).

Again due to time constraints, only one acoustic calibration date (August 11, 2015) was analyzed. On this date, two of the recordings for the ewes in the open pasture system were invalid because the microphones were not properly seated. Out of the 28 prehension events counted manually from the video files in a given segment of time, 26 were also detected automatically by GRASS (Table 3). As might be expected, start times of events were variable between the methodologies, ranging from 0.377 seconds to -0.124 seconds (Table 3, Figure 9).

The acoustic grazing data was also only analyzed for one date (August 4, 2015) and only one ewe per treatment within that date (r=3). From these preliminary results, the grazing time as determined by manual classification of the time-lapse images correlated well with prehension occurrences as determined by automatic detection within the acoustic files by the GRASS software (Figure 10). Also similar to the phenomenon observed in the time-lapse images, the acoustic results indicate that the ewe in the open pasture system spent the most time grazing compared to the ewes in the other systems. The ewe in the open pasture system also displayed a higher bite rate. However, ewe #1672 and ewe #1770 spent the least amount of time grazing and ewe #1654 spent the greatest amount of time grazing of the ewes within each of their respective experimental units (Figure 8), so it remains unclear how grazing time as detected by GRASS will change as more data is processed.


Participation Summary

Educational & Outreach Activities

Participation Summary:

Education/outreach description:

The methodologies utilized in this research project were demonstrated on site during a field day to the students of the 2015 Virginia Governor’s School for Agriculture. Some of the methodologies were also demonstrated and practiced by students in the Crop Management and Agroforestry Systems courses during two separate field days on site.

Components of this research project will be presented in a poster session at the American Society of Agronomy’s national meetings (November, 2015) and the American Forage and Grassland Council’s annual meeting (January, 2016).


Project Outcomes

Project outcomes:

This work has demonstrated the feasibility of using acoustic recordings and analysis to compute grazing dynamics of free range sheep. Preliminary results indicate that prehensive event occurrence, including bite rates, can be accurately determined through wideband acoustic recordings.

This work has also demonstrated the feasibility of implanting small temperature loggers into blank CIDR devices for monitoring core temperatures of lambs in free range settings. Though the results still require statistical clarification, this system provides an accurate and low-interference methodology for studying shade effects on sheep physiology.

Though GPS did not prove accurate enough in this research project, we were able to demonstrate the suitability of time-lapse cameras for documenting lamb behavior and shade utilization. This methodology provides confirmation of the accuracy of the acoustic analysis system and provides information that can be correlated with the core temperatures of the lambs to determine the impact of shade use on lamb physiology.

Though data analysis remains to be completed, it appears that different tree species in silvopasture systems have variable effects on forage productivity, lamb performance, and lamb behavior.

It is expected that this research project will be included in the graduate student’s dissertation. Likewise, it is expected that this work will be published in scientific journals and presented at agronomy and forage conferences. As more information is gained on how tree species impacts nutrition and behavior of lambs, this information will be shared in extension publications and field days to help producers make silvopasture implementation or management decisions depending on their desired objectives.



Areas needing additional study

A complete analysis of the collected data remains to be finished, including processing the forage nutritive value information. Another iteration of this summer grazing study will take place (2016). The results determined from the complete analysis of the data after that point will provide information on future directions for research.

Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture or SARE.