Measuring animal activity from signal strength recordings

 

William W. Cochran, Sparrow Systems, Fisher IL

Telephone: 217 897 1913  Email: sparrow@springnet1.com  

 

Contents

 

Background and Quick Primer

 

Introduction

 

Methods, theory

  Radio Equipment

  Animal movement and signal change (Ds)

  Visually derived  estimates of  activity

  Reducing sequences of signals (S) to digitized states via DS.

  Converting state data to quantitative activity indices (A%)

  Accuracy of A% estimates

  Processing state data to quantify clumped activity

 

Practical Matters

  Interference, false or fragmented data

  Site selection

  Relations among Dt, SI,  the number of animals (N), and transmitter specifications

  Getting A% while direction finding

  Comparison with activity switch methods

 

Glossary of terms

 

Literature & Bibliography

 

Bibliography

 

Appendix I: ARU technical
Appendix II: Worklists, Number of Channels
Appendix III: Additional figures
Appendix IV: Viewing and analysis tools
Appendix V: Review of methods literature

Background and Quick Primer

            If you don’t have time for details, jump over Background to Quick Primer and stop at its end..

 

Background

            Development of the method to be described began in earnest about 1990 although I developed equipment for the application much earlier, ie.,  the Falcon Five tracking receiver and accessories in 1977-1982, marketed by Wildlife Materials, Carbondale IL.  Development of the earliest fully computerized receivers (computer and receiver built as a single unit) was partially sponsored by the National Audubon Society for their California Condor project (1981-82). In 1990 the Condor project design was improved and a number of units built. I used these to conduct studies for the Illinois Natural History Survey (INHS) and, after retirement (1992), for myself. Some of the data presented here are from this 1990-93 period. From 1995-98, employed by the University of Illinois ECE department, I and colleague George Swenson continued using the equipment with support from the U.S Army Construction Engineering Research Laboratory (CERL, Dr. Larry Pater). Much of the data presented here is from that era. In 1998, under contract with CERL, further modification of the automatic receiver was made, mainly to vastly increase its memory storage capacity; in 2001 CERL funded development of a smaller and final version of the automated receiver, one that would cover a 20 MHz frequency range as opposed to the 1 MHz coverage of previous versions. The contract developments were made by myself and my son James through his company, Sparrow Systems, now manufacturing the final version. The data presented here from 1999 to present were obtained with the prototypes or final version.  Hopefully, other data will soon appear in print from a CERL desert tortoise project and various projects at the Smithsonian Tropical Research Institute (STRI) in Panama; these projects employ a number of the automatic receiving units (ARU) supplied by Sparrow Systems.

            Arlo Raim, from 1992 to present, often as a volunteer and variously employed by the UI ECE dept, and INHS,  is and has been a key participant in operating the equipment and gathering much of the data presented. Thanks are due and here given to the National Audubon Society, the INHS, and the UI ECE department for their support, interest, or benign consent during the early development period. Special thanks are due Dr. Larry Pater and CERL;  Pater’s vision, enthusiasm, and drive and the contracts from CERL made possible the conversion of  ad-hoc prototypes into a commercially available product (ARU) that we hope will find use in the wildlife ecology field. Feedback from Larry Pater, George Swenson, Dave Delaney, and Andrew Walde (CERL project) and Martin Wikelski, Roland Kays, and Franz Kuemmeth (STRI  project) is appreciated; it has helped reveal bugs in and fine tune the various programs used in running the ARUs. It should be noted that the CERL and STRI projects employ the equipment for direction finding and radiolocation by triangulation as well as for generating activity data; only the latter is the subject matter here.    

 

Quick Primer

            Aural or visual detection of signal change has been used in a number of studies ( 3, 4, 6, 8, 10, 11, 15, 17, 19, 23, 26, 34, 36, 43, 44, 47, 52 ) and in the era of paper and tape recorders, it was the only method available. These approaches adequately revealed the presence of or times of onset and cessation of activity in relation to environmental factors such as change in food supply, ambient light level, temperature, etc. Such timing may be all that is asked of data, with no attempt at a quantitative estimate of the level of activity in the intervals between activity onset and cessation.  

            A quantitative employment of the amplitude change method is based on body movements and posture changes that cause a change in received signal. Animal posture changes cause variations of proximity of an attached transmitter and its antenna to the animal, causing the radiated power to vary due to resultant changes in antenna efficiency and impedance. Also, because the signal radiated in different directions varies with the transmitter antenna orientation, movements of an animal as it turns or raises or lowers its head cause the power radiated toward the receiver to vary. The combination of varying degrees of change of posture and orientation result in a chaotic relationship between signal change (DS) and movement; small body movements sometimes causing large change and vice versa. Change, therefore, is not a valid measure of the movement that caused it. It is most important, however, that there is virtually no change when there is no movement,  permitting successive pairs of signal measurements to be categorized either as no change = no movement or some change = some unknown amount of movement.

            It is the percent change in signal that needs to be recorded, not the absolute change. Typical wildlife receivers provide the latter, which has led users to believe that change is not a reliable indicator of movement (trace this false rumor back from ref 25). A receiver with output (S) proportional to the logarithm of input automatically provides the needed percentage signal changes that are here called DS. In addition, it does not saturate at high signal levels (animal nearby) or lose resolution at low levels (animal distant). The Automatic Receiving Units (ARU) that we used record S values for many animals at set time intervals (Dt, seconds).

            The signal change DS between successive S measurements S1 and S2 spaced by time interval Dt was tested against a change threshold T1 (Figure 3). Two states may result, active (A) when DS => T1 and inactive (I) when DS < T1.  As noted above, DS is not a measure of the amount of movement during Dt, instead, it is an indicator of whether some unknown amount of movement has (DS => T1) or has not (DS < T1) occurred during Dt. Thus, a count of A states over a sampled interval (SI) is not a count of movements and, as we show later, derivation of a quantitative index of activity (A%) must derive from a substantial collection of state data (an SI much much greater than Dt).

            A minimum signal threshold T2 is necessary to account for periods when the signal is very weak or absent, such as when the transmitter is distant or underground. Signal cancellation from multipath propagation or signal nulls due to transmitter antenna orientation can also cause very weak or undetectable signals, even when the average received signal is strong. Occasional and extended periods of very weak or absent signals are handled by application of T2 as follows. If S1 < T2 and DS => T1 an A state results. If S2 < T2 and DS < T1 an I state results. If S1 and S2 are both < T2, then a third weak-signal (W) state results.

            In summary, DS measurements are evaluated as above to create a temporal data set of A, I, and W states. We refer to temporal plots of state data as actograms (14. Georgii 1981, Figure 4). Actograms provide an excellent overview of the temporal distribution of activity but differences in the intensity of activity, say for whole days from one day to another or one animal to another, are not visually apparent unless they are large (Figure 5).

Quantified differences in intensity of activity for different weather, seasons, climates, habitats, individuals, sexes, ages, etc. could, it seems, enhance understanding of the general ecology of animals. Actograms provide an excellent visual impression of differences, but impressions are subjective; a quantitative measure is preferable, such as that shown in Figure 6.

            Activity indices (A%) were derived as follows. Over an SI  >> Dt, a state ratio (the sum of A states divided by the sum of A and I states) can be expressed as an activity percentage for that SI (A% = 100 x the state ratio). A% is a quantitative index to an average of an animal’s movements during SI. It is an index because the amount of movement represented by each state is unknown. It is a quantitative index because details of the random and somewhat chaotic relationship between movement and individual DS tend to average out over SI. We emphasize that A% is an index derived from an average of many DS measurements that are randomly related to but on average positively correlated with the movements that cause them. As an average over SI,  higher accuracy of A% would be expected for longer SI and shorter Dt, ie., for larger sample sizes (more states). We found this to be true (Figure 10); for typical values see Table 1.

                Time plots of S show periods of inactivity interspersed with periods of activity (Figures 1, 2). The state data were processed to detect and measure intervals over which I states outnumber A states and vice versa. We call these intervals sessions, either inactive or active; a session is dominantly rather than exclusively one or the other state (see Figure 12a). Frequency distributions of the session durations (visible in the masked actograms Figure 11) can be extracted to obtain another perspective of animal activity behavior (Figure 12b). The intensity of individual sessions (Figure 12a) may also provide useful information.

 

                                    Introduction

 

Early in 1962, shortly after animal telemetry appeared,  Professor Dwain Warner (1963) articulated his vision of a need for and value of motion studies.

“One of the most significant observations of an animal is its motility. Since the environment apparently furnishes or modifies the stimuli which trigger basic responses within an animal, the ability to follow an animal’s movements continuously and to attempt to correlate these movements with environmental factors seems essential to advances in ecology. Nearly all animals are at some time motile;  and, since the causal factors of this motivation in the natural environment are not adequately understood, continuous recording of animal movements correlated with other events in the environment should be given major emphasis .... The need to know an animal’s diel (24 hr), seasonal, and annual movements and the physical as well as biological stimuli responsible for these movements has become paramount ....  

Warner’s reference to movements necessarily included locomotor activity as one of many  kinds of movement.

            Subsequently, radio location became a common method for tracking place to place movements of animals to determine their home range or territorial behavior. In many of these tracking studies, it was noted that bearing measurement was often difficult due to random short-term signal pitch and/or amplitude changes caused by animal body movements, including place to place flying, walking, and running. These signal changes were sometimes recorded as indices of motility from which diel and other patterns of activity were derived  ( 1-8, 10, 11, 13, 15, 17-19, 20, 23, 26-32, 34, 36, 40, 41, 43, 44, 46-50, 52, 53 ). No standard activity measurement methodology has emerged from these studies and virtually all used linear or qualitative non-linear detection that precluded quantitative analyses that would allow detailed comparison of results. However, these studies adequately show transitions between periods of little or no activity and periods of substantial activity, such as occur with the daily onset and cessation of activity.

            Limitations imposed by the processing and storage capacity of strip-chart or tape recorders vanished with the plethora of computers, making procurement and processing of large data sets practical. Large data sets, coupled with standardized quantitative analyses, provide an opportunity to evaluate overall animal activity to ascertain differences influenced by climate, season, habitat, habitat manipulation, etc. The simplest categorization of activity into 2 states (has moved or has not moved) falls short of Warner’s ideal that would include the nature of these movements and their function. Temporal patterns and prior knowledge of behavior can sometimes enhance interpretation of results. References 1, 3, 9, 15, 18, 21, 33, 36, 43, 49 discuss some aspects of 2-state activity analyses, similar to One-Zero state sampling used in visual behavior studies of primates ( 54. Altmann and Wagner, 1970).

In this manual we describe an activity monitoring method that uses signal changes, but is immune to signal strength bias and has accuracy equivalent to methods using transmitters with motion sensitive switches. The method will work with conventional wildlife transmitters, pulsed or continuous, and in addition, is more flexible in application than motion switch methods and can be implemented in an ongoing study with tagged animals already in the field. Data for several different species are used to illustrate the method. We describe equipment for logarithmic detection of signals, algorithms for standardized computer processing of signal strength data into activity indices, methods for estimating the accuracy of the activity indices, and a method for producing frequency distributions of the duration of activity sessions (bouts). A brief comparison with switch sensor activity monitoring is given.

                                    Methods and Theory

Radio Equipment

The data we present here were collected by automated receivers operated by built-in microcontrollers. These automated receivers (Model ARU, Sparrow Systems, Fisher, IL) are portable (15 x 15 x 15 cm weighing 1000 g); a power consumption of  0.2 watts permits operation from an automobile battery for 2 months, or indefinitely if charged from a 5 watt solar panel. Receiver output is proportional to the logarithm of signal input over a dynamic range of 75 dB (–140 to –65 dBm). Automation includes the agility to sequence through a user selected list of up to hundreds of frequencies in the 148 to 170 MHz range. Pulse amplitude to the nearest 0.3 dB, pulse width and interval to the nearest 0.001 second, and date-time to the nearest second were recorded on exchangeable memory modules with one month storage capacity. In some instances these data were transmitted via a wireless serial link to a central location for display and analysis with memory module storage used as a back up. For activity-only studies, a single omni-directional antenna was used; for direction finding, we used circular arrays of directional antennas (Larkin et al 1996, Cochran et al 2001).

The automated receivers briefly described above (more details in Appendix I) are convenient but not unique; any receiver that can be calibrated in dB can be combined with a computer to provide equivalent signal processing, frequency sequencing, antenna control, and data storage. However, if portability and power consumption are important factors, they should not be overlooked in the process of equipment selection.

 

Animal movement and signal change (DS)

The amplitude change method is based on body movements and posture changes that cause a change in received signal (Rin). As an animal moves from place to place, changed distance to the receiver will cause large changes in the average Rin. This is important because received signal change for a given body movement is greater for strong Rin from nearby than for weak Rin from a distance. For example, a head-down to head-up movement might cause Rin to double, and if the animal was nearby, Rin would double from say 100 to 200 mW (DRin = 100 mW), whereas if the animal was more distant, Rin could go from 0.01 mW to 0.02 mW (DRin = 0.01 mW). A linear receiver would output these two Rin, one 10,000 times greater than the other, both representing the same body movement and percentage change in Rin.

Logarithmic detection (Cochran, 1980), where receiver output (S) is proportional to the logarithm of Rin, eliminates the distance bias because changes of S (DS) are proportional to the ratio of the Rin that constitute the change; i.e., the percent change in Rin (Figure 1a).  DS (S1 – S2) is conventionally expressed in dB = 10 x  log (Rin1 / Rin2) where Rin1 / Rin2 is a power ratio. In addition to providing DS proportional to percent change of Rin, a receiver with a logarithmic response does not saturate at high Rin levels (animal nearby) or lose resolution at low Rin levels (animal distant), ie., over more than a million to one (60 dB) range of Rin, DS maintains a constant resolution for percentage changes in Rin. Although a linear-to-log transformation of the output of a linear receiver would remove a distance bias, linear receivers are capable of handling an Rin range of only a few hundred to one (26 dB) before saturation. 9 dB is a typical S change for doubling distance so 26 dB would represent about a 1-to-8 distance ratio (say 100 to 800 meters) whereas 60 dB would provide more than a 1-to-64 distance ratio (say 50 to 3200 meters). Moreover, since coverage area is proportional to distance squared, the value of a high dynamic range is greater than these distance figures suggest.   

Animal posture changes cause variations of proximity of an attached transmitter and its antenna to the animal, causing the radiated power to vary due to resultant changes in antenna efficiency and impedance. Also, because the signal radiated in different directions varies with the transmitter antenna orientation, movements of an animal as it turns or raises or lowers its head cause the power radiated toward the receiver to vary. The combination of varying degrees of change of posture and orientation result in a chaotic relationship between DS and movement; small body movements sometimes causing large DS and vice versa. Therefore, with one exception, the magnitude of DS is not a valid measure of the movement that caused it. The exception, that a very small DS implies no movement, is an important one because it permits each DS to be categorized either as some unknown amount of movement or as no movement.

 

Visually derived  estimates of  activity

The times of onset or cessation of activity in relation to environmental factors, such as change in ambient light level or temperature, may be all that is asked of the data, with no interest in levels of activity before or after the change. The times of changes may be taken from the time axis of S plots, for instance the end activity of nine Quail in Figure 1a occurred within a minute or two of 1958 CST for all of them. The timing of such transitions is usually clear in plots because S changes (DS) for no movement are much smaller than those produced by even moderate movement. Carried a step further, the start and end times of any periods of activity may be read from S plots and the durations of these periods used to estimate activity as, for instance, a percentage of a day ( Figure 1b). Visual selection becomes difficult and more subjective when S plots contain a many short periods of inactivity (Figure 2a), but this selection can to some extent be automated (Figure 2b).

 Visual analysis has been used in a number of studies ( 3, 4, 6, 8, 10, 11, 15, 17, 19, 23, 26, 34, 36, 43, 44, 47, 52 ) and in the era of paper chart recorders, it was the only method available. As many of the figures presented here show, computer processing of ARU data can provide a variety of useful visual presentations from which activity estimates and other insights into animal behavior obtained. However, with data already filed in digital form, exploitation of opportunities for more detailed analyses is logical.      

 

Reducing sequences of signals (S) to digitized  states  via DS.

            The magnitude of signal change DS between successive S measurements S1 and S2 spaced by time interval Dt was tested against a change threshold T1 (Figure 3). Two states may result, active (A) when DS => T1 and inactive (I) when DS < T1.  As noted above, DS is not a measure of the amount of movement during Dt, instead, it is an indicator of whether some unknown amount of movement has (DS => T1) or has not (DS < T1) occurred during Dt. Thus, a count of A states over a sampled interval (SI) is not a count of movements and, as we show later, derivation of a quantitative index of activity must derive from a substantial collection of state data (an SI much much greater than Dt).

A minimum signal threshold T2 is necessary to account for periods when the signal is very weak or absent, such as when the transmitter is distant or underground. Signal cancellation from multipath propagation or signal nulls due to transmitter antenna orientation can also cause very weak or undetectable signals, even when the average received signal is strong. Occasional and extended periods of very weak or absent signals are handled by application of T2 as follows. If S1 < T2 and DS => T1 an A state results. If S2 < T2 and DS < T1 an I state results. If S1 and S2 are both < T2, then a third weak-signal (W) state results. T2 may be set at any value, but high values will result in unnecessary loss of data, ie., many W states when A or I states could be confidently assigned. Setting T2 low, however, will cause periods of very weak or absent signals to produce A and some I states as variation in S measurements of the fluctuating noise floor randomly mimic activity or inactivity. A practical value for T2 is  -128 dBm, 4 dB above the average S value caused by noise alone (about  -132 dBm for a receiver bandwidth of 2 KHz and a typical rural external noise environment). To preserve temporal continuity in the data set,  no-data (Z) states are inserted for periods when the equipment is not operating.

            In summary, DS measurements are evaluated as above to create a temporal data set of A, I, and W states. We refer to temporal plots of state data as actograms (14. Georgii 1981, Figure 4). Actograms provide an excellent overview of the temporal distribution of activity but differences in the intensity of activity, say for whole days from one day to another or one animal to another, are not visually apparent unless they are large (Figure 5).

 

State-derived activity indices (A%)

Quantified differences in intensity of activity for different weather, seasons, climates, habitats, individuals, sexes, ages, etc. would enhance understanding of the general ecology of animals. Actograms provide an excellent visual impression of differences, but impressions are subjective; a quantitative measure is preferable, such as that shown in Figure 6a 6b.

Over an SI  >> Dt, a state ratio (the sum of A states divided by the sum of A and I states) can be expressed as an activity percentage (A%) for that SI. A% is a quantitative index to an average of an animal’s movements during SI. It is an index because the amount of movement represented by each state is unknown. It is a quantitative index because details of the random and somewhat chaotic relationship between movement and individual DS tend to average out over SI. Accuracy and temporal resolution of A% compete via Dt and SI as will be demonstrated later.

A% is influenced by the selected T1 and Dt. In the extreme, as T1 is made large or Dt short, the probability of DS causing A states approaches zero, biasing results toward the I state and reducing A%. Conversely, as Dt is made long or T1 small, the probability of I states is reduced, biasing results toward the A state and increasing A% (Figure_7a_7b).

It is axiomatic that as Dt approaches zero, DS and A% must also approach zero.  The effect of long Dt is less clear. The proposition that it is more probable that an animal will have moved at least once over a long time interval than over a short one is sound. However, the signal may have changed many times during a long (or any) interval and at the instant of measurement differ by less than T1 from the previous measurement. In other words, a false I state will sometimes be generated no matter how long the time interval and no matter how much movement has occurred during the interval. These false I states prevent A% from reaching 100%, even for a very active animal and a long Dt. (Figure 7a 7b). Of course three consecutive A states (or any  small number of consecutive A states) make A% equal to100%, but such small samples do not adhere to the rule that SI >> Dt.

 

Accuracy of state-derived estimates of  activity ( A% )

                We emphasize that A% is an index derived from an average of many DS measurements that are randomly related to, but on average positively correlated with, the movements that cause them. As an average over SI,  higher accuracy of an A% index (100 * A / (A + I) ) would be expected for longer SI and shorter Dt, ie., for larger sample sizes (more states). We found this to be true with tests as follows.

To test accuracy, S readings of a radio tagged Swainson’s Thrush were made at 0.5s intervals for 9 days (Figure 8). On each of the days the bird was inactive from 2000 to 0500 the following morning. We sampled from this 9-day data set with various Dt and SI, using multiple passes at identical Dt. For example,  to estimate the accuracy for Dt = 60s for a days activity (SI = 15 hours from 0500 to 2000, 900 S reading), an A% was calculated from every 120th S reading, starting with the 1st reading. This was repeated starting with the 2nd reading, then with the 3rd, and finally with the 120th.  Thus A% was calculated 120 times from the same data set with each pass through the data starting a half second later than the previous. The 120 data sets covered virtually the same period of time: 0500:00 to 1959:00 for the 1st and 0500:59 to 2000:00 for the 120th. The above procedure was repeated for each of the nine days, providing nine SD estimates of the error for a Dt of 60s and an SI of 15 hours of the bird’s active period; The 9 SD ranged from 1.18 to 1.49  (mean 1.38, SD 0.084) (Figure 8bc). It was repeated again for an SI that included the inactive period (24 hours); the 9 SD ranging from 0.74 to 0.93 (mean 0.85, SD 0.054). The strong correlations of Figure 9b evolve because the added I states increase sample size (SI) without adding variability; this holds regardless of SI or Dt (see discussion of Figure 9 below). In other words, the ratios of A% values without and with inactivity periods included, and their respective SD values, are 1.6:1, which is the ratio (24:15) of SI with and SI without the inactive periods (Figure 9b).

If the method was perfect, each day’s 120 data sets would produce virtually the same A% for that day because they sample almost identical time periods in the same data set. Not being a perfect method, A% varies from pass to pass, and these A% variations are a measure of error of the method. The frequency distributions of A% (Figure 8b) are close to normal, thus standard deviation (SD) is an appropriate error statistic.

The procedure in the above example was repeated for other Dt (7.5, 15, 30, 120, and 240s) and other SI (3 days = 45 hours and 9 days = 135 hours) producing the upper plots of Figure 9a. Similarly, the procedure was repeated with night inactivity included (SI = 24, 72, and 216 hours) to produce the lower plots of Figure 9a. It was no surprise that including night inactivity reduced SD. Intuitively, if A% = 0 (all I states) there should be no variability (zero SD and zero error). It follows that data sets that include a large number of consecutive I states will produce a lower SD than those that include them, as is shown in the upper and lower plots in Figure 9a also see Fig 2b. The only measure of inactivity is duration, and if this alone is sought, it may be taken directly from S plots or actograms (or see Processing state data to quantify clumped activity section below). We here speak of inactivity as it occurs over long intervals of say rest or sleep, not of short bouts of less than 15 minutes or so.

Periods of substantial activity, easily seen in S plots and actograms, have two measures, one of which is their duration, as with inactivity. The other is the intensity of activity during the periods, such activity consisting of a mixture of varying short periods of activity and inactivity. The two measures combine as a series of products (duration times intensity) which along with long periods of inactivity yield a movement index to activity (A%) over a sampled interval (SI) such as an hour, day, week, or season. Thus, long periods of inactivity are reasonably included in A% calculations. The accuracy of these A% calculations that include all data, however, is erroneously high (SD low) as we have shown.

Fortunately, inclusion of long periods of inactivity reduces calculated SD in a simple way, ie., in proportion to the ratio of their duration to SI. For example, if SD is calculated for a day of data and the animal sleeps 11 hours at night and takes a 1-hour noon nap, use twice the calculated SD (24/12) as the basis for accuracy of the day’s A%. If an animal has no clear day (or night) period of inactivity, the sum of the longer and obvious periods of inactivity may be used, for instance the long periods easily seen in Figure 1b, but a better measure of them is done with a program (see Figure 2b) .

The data from Figure 9a, along with additional data for SI = 3.75 and 7.5 hours (during the active period), were combined to produce Figure 10a from which a desired accuracy (a user selected function of SD on the Y axis) can be related to SI (temporal resolution, on the X axis) and Dt (plotted curves). The equation SD = SI -0.513 (Figure 10a) illustrates the classical relationship that accuracy (1 / SD) is proportional to the square root of SI (sample size).

The discussion above, revolving around Figures 8 9 10a, is based on processing with T1 = 4 dB. It is evident from Figures 7a 7b that A% increases substantially with decreasing T1. SD for a given Dt, however, is little affected by T1 (Figure 10b). We have also focused explanations and examples around a particular Swainson’s thrush for which S readings were taken at 0.5s intervals (this is Swainson’s thrush 2005 in Table 1). As is evident in Table 1, data taken at 10s or 15s intervals may be adequate for estimating SD if SI is long (many days); the SD shown in the Table are valid for Dt = 60s, ie., it would be reasonable to consider these data as focal data that provide a Dt = 60s SD and to plan to monitor many animals with S readings taken at 60s intervals (see the section: Relations among Dt, SI,  the number of animals (N), and transmitter specifications for the importance of this). If a short Dt (<< 60s) is practicable in a study, processing for A% can be done with Dt = 60s by skip sampling while continual estimates of SD are made from the 4 or more samples of A% calculated from multiple passes; in this case note that the basis for estimated error is the calculated SD divided by the square root of the number of passes, eg. for the data in the Table, divide the SD by the square root of 4 (or for the Red Fox, the square root of 6). In all this discussion, we focus on 60s or 15s only as a means of example, but there is nothing magic about these intervals; other numbers may be plugged in and the principles still apply (see Fig 10). We have also provided SD numbers to 2 decimal places, which is almost silly. In practice, selecting the SD used as basis for error estimation would require looking at data from several animals, perhaps in each season or environmental situation or behavioral state, and making a judgement. For instance the Quail in Table 1 were lone survivors (not in a covey) and it was summer; all their SD but one were in the range 1.5 to 1.65.                 

 


Table 1.  Standard deviations (SD) of error of A% for several species, calculated for active-only periods and for 24-hour periods. For each day of the sample, four A% values were calculated using S readings at 15s intervals and the SD of these four A% was calculated. The SD (of A%) shown are for Dt = 60s; they are the means of N of these 4-pass SDs (N = number of days); and variation around the means is indicated by SD of SD. The Red Fox is an exception; there were six A% values for each day taken at 10s intervals (other plots of its data are in figures 1a, 6b, and 11b. Eastern bluebird data appear in figures 11a, 12, and 14, and data for the Northern cardinal appear in figure 11d. Other data for these animals are in Appendix III.

 

Animal

    Dates

 

Days

   Active only for SI hours

  For 24 hours

 

 begin

 end

 

SD

SD of SD

SI hours

SD

SD of SD

Rabbit 117

17-Nov

9-Dec

22

1.04

0.58

12.0

0.84

0.39

Rabbit 157

30-Jan

2-Mar

22

1.55

0.65

13.0

0.90

0.38

Rabbit 220

8-Dec

31-Dec

21

1.22

0.76

14

1.01

0.48

 

2-Jan

24-Jan

21

1.44

0.58

14

1.04

0.50

 

 

 

 

 

 

 

 

 

Red Fox adult

31-Aug

7-Sep

8  

 

 

 

1.04

0.51

 

17-Sep

28-Sep

12

 

 

 

1.02

0.50

 

10-Oct

31-Oct

15

 

 

 

0.95

0.54

 

 

 

 

 

 

 

 

 

Downy Woodpecker 1

29-Jun

9-Jul

10

1.50

0.62

10.0

1.02

0.48

Downy Woodpecker 5

30-May

21-Jun

23 

1.53

0.72

12.0

0.75

0.35

 

 

 

 

 

 

 

 

 

Swainson's thrush 2005

1-Jun

10-Jun

9  

1.39

0.74

15.0

0.86

0.47

Swainson's thrush 188

21-Sep

1-Oct

11

1.60

0.67

11.0

0.92

0.43

Fox sparrow

17-Oct

29-Oct

8  

2.20

1.40

11.0

1.22

0.72

 

 

 

 

 

 

 

 

 

Eastern bluebird

9-Jul

16-Jul

7  

1.21

0.52

15.5

0.83

0.38

 

17-Jul

23-Jul

7  

1.48

0.67

15.5

1.02

0.44

 

 

 

 

 

 

 

 

 

Northern cardinal

12-Aug

12-Sep

32  

1.52

0.71

12.5

0.92

0.41

 

 

 

 

 

 

 

 

 

Brown-headed cowbird

26-Jun

4-Jul

9  

1.52

0.75

16.0

0.90

0.43

 

 

 

 

 

 

 

 

 

Bobwhite quail 328

5-Jun

12-Jun

8

1.10

0.54

15.0

0.78

0.36

Bobwhite quail 347

5-Jun

12-Jun

8

1.52

0.63

15.0

1.00

0.45

Bobwhite quail 399

5-Jun

12-Jun

8

1.45

0.58

15.0

0.98

0.40

Bobwhite quail 524

5-Jun

12-Jun

8

1.63

0.67

15.0

1.07

0.48

Bobwhite quail 555

5-Jun

12-Jun

8

1.61

0.77

15.0

1.11

0.51

 

 

 

 

 

 

 

 

 

 

In summary, SD can be calculated from a number of A% values from multiple passes with different start times through the same data. Frequency distributions of these A% were found to be approximately normal, making their SD a good basis for calculating A% error probabilities. Data sets with long periods (say >15 minutes) of inactivity removed provide the best estimate of SD for A% of the same data sets with no data removed. SD was found to be little affected by T1 over its typically useful range of 3 to 5 dB. SD was found to be approximately inversely proportional to the square root of N (SI / Dt), for example, the significance of differences of A% from week to week will be greater than that from day to day.

 

Processing state data to quantify clumped activity

Visual identification of clumps (sustained periods) of inactivity from S plots (Visually derived estimates of activity section) is a simple way to estimate a day’s activity as the ratio of the minutes in a day minus the sum of the clumps, all divided by the number of minutes in a day and finally multiplying by 100 to convert to percentage. In other words, having identified clumps of inactivity, the remainder must be clumps of activity. The clumps of inactivity can be identified quantitatively from state data as periods of consecutive I states longer than a user selected number of minutes, say 15 minutes as in Figure 2b where these clumps are called I bouts. This method is somewhat erratic because single A states sometimes break a sequence that obviously qualifies as a sustained period of inactivity, eg. from 0540 to 0600 in the upper S plot in Figure 2b.

            We here present a majority logic method that is not erratic because instead of detecting unbroken sequences of I states, it detects intervals over which I states outnumber A states and vice versa. We refer to these intervals as inactive (Si) and active (Sa) sessions where a session is dominantly rather than exclusively one or the other state.

The process is one of filtering out the minority states. Quantitative removal of minority states is low-pass filtering, a process that creates from a sequence of state data a second sequence of filtered state data for each time (t) in the data set. For instance, an 11-state data nibble could consist of 9 A states, begin and end with an A state and contain 2 I states. Majority thought and logic would consider this 11-state nibble to represent primarily activity rather than to consider it as say separate 4-state and 5-state nibbles plus two 1-state nibbles.  Thus considered, the mid point of the 11-state nibble (at time t) is set as A.  Time t is then moved forward to t + Dt and the process repeated with the oldest state (at t – 5 Dt) dropped from the nibble and the new state (at t + Dt + 5 Dt) added to maintain an 11-state nibble from which the state to be assigned to t + Dt is determined. The process, continued throughout the desired data set, produces a filtered set of state data with majority states emphasized. In this example for time t, all states, regardless of their time differences from t, were implied to have equal weight (1 vote) in the majority decision, ie. a linear filter. In a triangular filter, the state at time t would be given 6 times as much weight as the states at t + 5Dt and t - 5Dt, 5 times the weight of those at t + 4Dt and t - 4Dt, etc.

The above digital low-pass filtering is (mathematically) the passing of a symmetrical convolution mask over the state data. The width of a mask is usually referred to as a (time) window (Wt) that is an odd number of Dt's in length. In the preceding word example the time window, called a nibble, was 11Dt (Wt = 11Dt). Each state datum in the window is assigned plus 1 for an A  state,  minus 1 for an I state, and zero for a W state. At each time (t), the products of the mask and window of state data centered on time t is summed and the t assigned as A or I when the sum was > 0 or < 0, respectively (Figure 11, 12a). We found triangular masks, (eg., 123454321, Wt = 9) more effective than linear masks with equivalent high frequency cut-off  (eg. 1111111, Wt = 7) because the latter tend to create some detail where none exists. Digital filtering and a variety of masks are well covered in the literature (see 55. Chi-Tsong, 1979; chapter 7).

Ties (mask sum = 0) are infrequent but happen. A simple tie breaker rule is to favor the sign of the preceding (t -1) sum or to widen the mask symmetrically until the tie is broken in favor of the overall context. A few W states (assigned value 0) do not invalidate the process; too many weaken the decision. When W states were in the majority a W state was assigned and plotted in the session actogram. Although plotted in Figures 11 and 12b, the durations of active and inactive sessions adjacent to a session of no-data or weak signal  was not included in the file of session durations, ie., active and inactive sessions that began with or ended in a weak or no-data session were discarded for purposes of plotting session frequency distributions because their durations were unknown.

            Summarizing, state sequences can be low-pass filtered to visually reveal sessions (sometimes called clumps or bouts) of activity (or inactivity), defined as periods of time dominated by either activity (A states) or inactivity (I states). Low pass filtering may reveal patterns that might otherwise be lost in detail. Frequency distributions of the session durations (visible in the masked actograms Figure 11) can be extracted to obtain another perspective of animal activity behavior (Figure 12b). The intensity of individual bouts (Figure 12a) may also provide useful information.

 

                                    Practical Matters

 

In the preceding methods sections we have focused on principles illustrated with examples. Here we show how these principles apply in planning a study (Relations among Dt  SI  the number of animals (N) and transmitter specifications, Choosing transmitters, Getting A% while direction finding) and in addition cover practical matters that can impact a study (Interference, Site selection). A brief comparison is made with switch sensor methods (Comparison with activity switch methods).

 

Interference

            External factors can affect accuracy and create data gaps. Nearby thunderstorms produce static that shows up on records as varying signal. Power lines are also a source of noise, often intermittent due to variable wind and humidity. Siting of ARUs away from power lines usually helps. These noise sources are broadband, thus appearing on all frequencies. Programming the ARUs to receive signals on one or more non-animal frequencies facilitates cutting (changing to Z states) periods with interference; loss of data is unavoidable. See (Figures 13, 14, 15).

False activity can also be produced when wind-swayed vegetation causes variations in the received signal. This effect is somewhat reduced by siting the receiving antenna away from trees, or if in a woods, then above the canopy. A stationary reference transmitter in the study area will aid in detecting periods of false activity (Figure 16). Sections of records thus identified may be Z-coded as no-data in the data files. Use of a higher value for T1, say 5 or 6 dB, can place most wind-caused DS values occurring during inactive periods to be correctly recorded as I states. If, for any reason, most data are processed with T1 = 3 or 4 dB, the A% taken with the higher T1 must be adjusted upward from charts like those of Figures 7a, 7b to prevent bias in comparisons.

Various frequency-discrete signals from 2-way communications, data transmissions, and spurious radiation from high power television and FM radio transmitters are more difficult to eliminate. These, except for the latter, are usually very intermittent; all are plentiful in urban areas. Where their frequencies can be determined, they can be avoided by not using these frequencies for animal transmitters. These signals are usually heavily modulated and of durations often different from the pulse widths of animal transmitters. ARU detection algorithms provided flagging most signals of this type by the detecting these differences.

Many garage door openers and other remote control devices are a source of continuous broadband noise. Although radiation from such devices has been illegal for many decades in the USA, they nonetheless abound; in some urban neighborhoods it is almost impossible to find a place without interference from such devices. Fortunately, their interference is usually limited to a radius of < 100 m, allowing siting an ARU to avoid them. This kind of interference does not mimic a signal, it merely produces a background noise that reduces the distance at which animal transmitter signals can be usefully detected.

 

Site selection

 A decision that sometimes precedes a study is where to conduct it. Sometimes there are good reasons to conduct it in a particular place. In the section on interference, we describe four types of external interference that can affect accuracy and create data gaps; we also mention ways to mitigate these effects. Nothing is to be done about wind or lightning other than the procedures mentioned. However, the effects of urban-related interference can be avoided or minimized by choosing a site well away from towns, power lines, and TV-radio station towers. Where convenience alone is the reason for using a particular area that is also interference prone, the potential for data sets being fragmented or rendered useless should be avoided by finding a better area. Part-time tracking of animals by listening to signals from a radio is orders of magnitude less susceptible to interference than continuous automatic monitoring of signals for the purpose of obtaining the clean and unfragmented signal records that simplify data processing and produce accurate results.

If study objectives are minimal, for instance to measure the circadian pattern of activity without quantification of activity intensity during active periods, then the above cautions are not so important because accuracy issues are much more relaxed. Whatever the study objectives, a survey of the RF environments in the planned area will be useful in finding frequencies to be avoided. Such a survey is not very useful in projecting power line interference because power lines are often quiet but always give interference sooner or later.          

 

Relations among Dt, SI,  the number of animals (N), and transmitter specifications

            In general, design of a study will involve specifying 3 parameters: accuracy (1/SD), temporal resolution (SI), and a number of animals (N). These parameters are related: SD is approximately proportional to Dt / SI (Figure 10) and N is inversely proportional to Dt.  The latter stems from the fact that measurements are made sequentially and the dwell time “listening” for each transmitter must be at least as long as its pulse interval (Pint) to assure that a pulse will be present for measurement. The time between measurements for each transmitter (Dt) is thus at least as long as the sum of Pint  for N transmitters (Psum).

The effect of N on Dt can be controlled with Psum by appropriately specifying transmitter Pint. With Dt controlled in this way, SI can be specified. Thus, it appears that accuracy (1/SD), temporal resolution (SI), and a number of animals (N) can be set arbitrarily. But there are practical limits to the specification of Pint.   

A typical wildlife transmitter has a pulse interval of 1s and a pulse width (Pw) of 16 milliseconds (ms). To maintain transmitter longevity-to-weight ratio while decreasing Pint,  Pw can be shortened in the same proportion as Pint, maintaining the same duty cycle or, alternatively, peak pulse power of the transmitter can be reduced in the same proportion. There are practical limits. Pw may be reduced to 4 or 5 ms and peak pulse power by a factor of 2 or 3. Thus, without affecting the longevity or weight of transmitters, a Pw of 4ms and halving power would relieve specification of the combination of SD, SI, and N by a factor of 8 as compared with using typical wildlife transmitters. For example, for a given accuracy and temporal resolution, 8 times as many animals could be monitored, or for a given number of animals, better accuracy and temporal resolution could be obtained. It seems something has been gained for nothing. But some reception range is lost when Pw or pulse power are reduced; for the factor of 8 above perhaps to 0.85 that for the typical transmitter. Moreover, since area of coverage goes as the square of range, coverage would be reduced to about 0.72.

Sometimes transmitter longevity is twice (or more) than that really needed, the idea being that more than enough is better than enough, insurance so to speak. But halving longevity will allow halving Pint without any impact on range and this alone will double N.

These deviations from “typical” may not be welcomed by some transmitter suppliers because they may find it inconvenient to produce them, or simply think they know better than you what you need. If so, find a supplier that knows its business, one who is willing to let the dog wag its tail.

            In practice, because of attrition and logistical factors, the number of animals being monitored will vary. To maintain a minimum number of animals under observation for statistical purposes, an N somewhat larger than the minimum number should be used. Thus, as the number of animals waxes and wanes below N, Dt could also be allowed to vary by equating it to the Psum of the actual number of transmitters being monitored. A varying Dt not only means varying accuracy, but complicates analyses and data presentations as well. Keep the N and Dt constant to avoid these complications. A more detailed discussion of N,  Dt, and the programming of single and double work lists is given in Appendix II.

As a final caution, loose transmitter attachment can cause signal changes from movement to be different from changes with a firm attachment, also a concern with switch type sensors. Consistency of attachment is thus important if comparisons of activity among individuals is desired.

            In summary, one need not be too concerned about choosing Dt or transmitters unless high accuracy and temporal resolution are required for large numbers of animals. It may be satisfactory that 30 typical wildlife transmitters can be monitored at Dt of 30s for a SD of error about 1% for A% summations of about 1 day (SI = 1440 min).

           

Comparison with activity switch methods

Having to think about the several factors involved with choosing Dt may seem an undesirable complication of the amplitude-change method. The same complication exists for switch type sensors whose reset time and angular and/or acceleration sensitivity are equivalent to Dt and T1, respectively. A disadvantage of switch type sensors is that Dt and T1 are set by the radio package design and the way it is mounted on an animal. Thus set in advance, no flexibility is allowed during data collection and analysis, and if the parameters turn out to be less than optimal, the researcher is faced with accepting this or redoing the study. In contrast, the amplitude change method allows the user to optimize Dt from samples taken early in a study and later select an optimum T1 during data analysis. Considerations for SI and N are virtually the same for both methods. Simple switch-sensor devices produce A and I state data similar to that from the amplitude-change method making analyses similar. More complex switch-sensor packages that store motion data or respond to specific motions will serve objectives not accessible with the simper switch devices or the amplitude change method. The ARUs we used are capable of recording pulse interval and pulse width or even a pulse code, allowing switch-sensor and signal-strength methods to be used simultaneously.

 

Getting A% while direction finding

The ARUs can be used for direction finding (DF) by switching among directive antennas (56. Larkin et al 1996, 57. Cochran et al 2001, ARTS Website). Between 2 and 6 pulses are required to determine a direction (Sparrow DF Manual in prep). The S data gathered in the DF process may be used directly to obtain A% indices. The penalty for such dual use is less accuracy or smaller N due to the longer Dt required while waiting for the additional pulses. In some installations for direction finding, the cost of a tower, weather enclosure, battery, and solar panel, in addition to the several antennas and an ARU,  may be large enough that addition of a second ARU and single antenna, specifically for activity, represents only a moderate percentage increase in cost. If N is large and activity is an important objective, this approach will avoid compromise.

 

                                    Glossary of terms

 

N                     Number of frequencies in a list that an ARU sequences through.

                        All frequencies need not be different, ie., the same frequency may

be in the list more than once, adding to N each time it appears in

the list. May be thought of as the number of animals monitored with

the understanding that animals monitored more than once in the

sequence increase N proportionally. For animals monitored once in

the sequence, the sequence repetition time (Dt) becomes the Dt for

processing their DS to obtain A%.

Pint                  Abbreviation for transmitter Pulse Interval in seconds. Pulse (beep)

                        rate is the reciprocal of Pint (beeps per second).

Psum                Sum of Pint for the various transmitters in the sequence (see N).

Pw                   Abbreviation for transmitter Pulse Width in milliseconds.

  

Session             A period of time dominated by a particular state. An active Session

is defined as a period dominated by A states, inactive by I states,

weak signal by W states, no-data by Z states. Clump and bout

may be used instead of session.

Clump              see Session

Bout                 see Session

 

Session             The degree of dominance of a particular state in a Session. The simple

intensity            majority rule is all or nothing (dominance); it ignores the popular vote,

                        which may vary from 51% to 100%; we call it intensity (of dominance). 

 

ms                    abbreviation for milliseconds (1 ms = 1/1000 sec)

 

RF                   Radio frequency

 

Rin                   Signal power present at an antenna input of an ARU, expressed

in milliwatts unless otherwise stated.  

 

Wt                   Width of convolution mask in Dt units. To convert to seconds,

                        multiply Wt by Dt. It is a time interval.

                       

nibble               A sequential set of states contained in the interval Wt.               

 

activity index     A% is an activity index, usually from the ratio of A states to the sum

of A and I states but can be a ratio derived from periods of activity and

inactivity determined visually from S plots or Actograms. However A%

is derived, it is an index because it is not a direct measure of movement.  

 

activity              Activity of an animal, moved (active) or did not move (inactive)

actogram          Digital (2-state) graphical illustration of activity and inactivity,

first mentioned in lit by Georgii

ARU                General term for Activity Recording Unit

A%                  An index to the percent of a sampled interval (SI) an animal was active

I%                    An index to the percent of a sampled interval (SI) an animal was inactive

                        I% is not actually used in the text because it is always 100 – A%, thus

stating A% determines I% automatically.

        

Dt                     Interval separating two S measurements from the same animal

dt                     Dt as in calculus as t approaches 0, not used except

in a few figures where a font problem prevented using D.

 

SI                    Sampled time interval >> Dt over which A% or I% is calculated

 

dBm                 decibels relative to 1 milliwatt, a measure of power

dB                    10 x log of a power ratio: 10 x log (s2 / s1) s unit watts

20 x log of a voltage ratio: 20 x log (s2 / s1) s unit volts               

 

s, s1, s2 etc      Signal received in watts or volts (must specify)

 

S plot               Time plot of S (signal) values

S                      A signal value expressed in dBm                      

10 x log (s / 1mw) expressed as dBm (dB relative to 1 mw)

where s is power in milliwatts (mw)

S1, S2 ..Sn       Sequential S measurements separated by interval Dt

 

DS                   ABS(S2 - S1), both expressed in dB relative to the same reference,

the reference is not important.

10 x log ( s1 / s2 ),  s1 / s2 is a power ratio

or    20 x log ( s1 / s2 ),  s1 / s2 is a voltage ratio

In all cases the sign is ignored and the function is made positive,

thus swapping s1 and s2, or S1 and S2, does not change the value.

dS                    as in calculus, rate of change of S as t approaches 0, not used except

                        in a few figures where a font problem prevented using D.

 

t                       General point in continuous time.

t1, t2, ... tn       Points in a sequence of times in Dt increments.

 

T1                    Threshold one, units dB, deltaS => T1 define an active datum,

DS < T1 an inactive datum.

T2                    Threshold two, units dBm, minimum S1 and S2 allowed in determining

a state as active A or inactive I. Not allowed if both S1 and S2 < T2.

State A             A state:  DS => T1 ,  active state.

State I              I state:   DS < T1,  inactive state.

State W            W state:   Unknown state, S1 AND S2 < T2.

State Z             Z state:   No-data state, no data gathered or invalid data edited out. 

 

 

                                    Literature cited

 

1. Ables, E. D. 1969. Activity studies of red foxes in southern Wisconsin. J. of Wildl. Manage. Vol. 33,  No. 1. pp.145-153.

 

2. Amstrup, S. C. and J. Beecham. 1976. Activity patterns of radio-collared black bears in Idaho. J. Wildl. Manage. 40(2):340-348.

 

3. Beier, P. and D. R. McCullough. 1988. Motion sensitive radio collars for estimating white-tailed deer activity. J. Wildlife Mgmt. 52(1). pp. 11-13.

 

4. Broekhuizen, S., Van't Hoff, C.A., Jansen, M. B., and F. J. J. Niewold. 1980. Application of radio tracking in wildlife research in the Netherlands. Pages 65-84  in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

6. Clapp, D. F.,  R. D. Clark, Jr., and J. S. Diana. 1990.  Range, activity, and habitat of large, free-ranging brown trout in a Michigan Stream. Transactions if the American Fisheries Society, 1990: 1022-1034.

 

7. Cochran, W. W. and R. D. Lord. 1963. A radio-tracking system for wild animals. J. Wildl. Manage. 27:9-24.

 

8. Cochran, W. W. 1980. Wildlife telemetry. Pages 507-520 in Sanford D. Schemnitz, ed., Wildlife management techniques manaul. Wildlife Society, Washington, D.C.

 

9. Cooper, H. M. and P. Charles-Dominique. 1985. A microcomputer data acquisition-telemetry system: a study of activity in the bat. J. Wildl. Manage: 49(4):850-854.

 

10. Custer, M. C., T. W. Custer, and D. W. Sparks. 1996. Radio telemetry documents 24-hour feeding activity of wintering lesser scaup. Wilson Bull. 103(3),  pp. 556-566.

 

11. Erlinge, S. 1980. Movements and daily activity pattern of radio tracked male Stoats, Mustela erminea. Pages 703-710 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

12. Fancy, S. G., Pank, I. F., Douglas, D. C.,  Curry, C. H.,  Garner, G. W.,  Amstrup, S. C., and W. L. Regelin. 1988. Satellite telemetry: a new tool for wildlife research and management. U. S. Fish and Wildlife. Service  Resource Publication No. 172.

 

13. Garshelis, D. L., Quigley, H. B., Villarrubia, C. R., and M. R. Pelton. 1982. Assessment of telemetric motion sensors for studies of activity. Can. J. Zool. 60: 1800-1805.

 

14. Georgii, Bertram. 1981. Activity patterns of female red deer (Cervus elaphus L.) in the alps. Oecologia. 49:127-136.

 

15. Gillingham, M. P. and F. R. Bunnel. 1985. Reliability of motion-sensitive radio collars for estimating activity of black-tailed deer. J. Wildlife Mgmt. 49(4) pp. 951-958.

signal change, reset, and tip,  levels, strip chart, visual, -, +, + 

 

16. Gillingham, M. P. and K. L. Parker. 1992. Simple timing device increases reliability of recording telemetric activity data. J. Wildl. Manage. 56(1):191-196.

 

17. Göransson, G. 1980. Animal activity recorded by radio tracking and an audio time lapse recorder. Pages 457-460 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

signal change amplitude, presence/absence, tape recorder, aural, +   

 

18. Green, R. A., and G. D. Bear. 1990. Seasonal cycles and daily activity patterns of Rocky Mountain Elk. Journal  of Wildlife Management. 54:272-279.

 

19. Hanley, T. A. 1982. Cervid activity patterns in relation to foraging constraints: western Washington. Northwest Science. 56:208-217.

 

20. Hardy, A. R. and K. D. Taylor. 1980. Radio tracking of Rattus norvegicus. Pages 657-665 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

21. Julien-Laferrière, D. 1997. The influence of moonlight on activity of woolly opossums (Caluromys philander). Journal of Mammology. 78:251-255.

 

22. Kitchens, J. A., Maier, H. A., and R. G. White. 1996. Effects of ambient temperature on activity monitors of radiocollars. Journal of Wildlife Management. 60:393-398.

 

23. Kjos, C. G. and W. W. Cochran. 1970. Activity of migrant thrushes as determined by radio-telemetry. Wilson Bull. Vol. 82, No. 2. pp 225-226.

 

24. Knowlton, F. F., Martin, P. E., and J. C. Haug. 1968.A telemetric monitor for determining animal activity. J. Wildl. Manage. 32(4): 943-948.

 

25. Kunkel, K. E., Chapman, R. C., Mech, L/ D., and E. M. Gese. 1991. Testing the Wildlink activity detection system on wolves and white-tailed deer. Canadian Journal of Zoology. 69(9): 2466-2469.

 

26. Lancia, R. A., Dodge, W.E., and J. S. Larson. 1980. Summer activity patterns of radio marked beaver. Pages 711-715 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

27. Lindzey, F. G. and E. C. Meslow. 1977. Home range and habitat use by black bears in southwestern Washington.  J. Wildl. Manage. 41(3): 413-425.

 

28. Loughlin, T. R. 1980. Radio telemetric determination of the 24-hour feeding activities of sea otters, Enhydra lutris. Pages 717-724 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

29. Lord, R.D., Bellrose, F. C., and W. W. Cochran, 1962. Radio telemetry of the respiration of a flying duck. Science., N. Y. 137, 39-40

 

30. Marshall, W. H. and J. J. Kupa. 1965. Development of radio-telemetry techniques for roughed grouse studies. Pages 443-456 in ed. J B. Trefethen. Transactions, 28th North

 

31. Marshall, W. H. 1965. Ruffed grouse behavior. BioScience. 15(2): 92-94.

 

32. Maurel, D. 1980. Home range and activity rhythm of adult male foxes during breeding season. Pages 697-702 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

33. Mohas, I. 1987. A storing telemetry-transmitter for recording bird activity. Ornis San. 18:3 pp. 227-230.

 

34. Nams, V. O., 1989. A technique to determine the behavior of a radio-tagged animal. Canadian Journal of Zoology. volume 67(2), 254-258.

 

35. Palomares, F. and M. Delibes. 1991. Assessing three methods to estimate daily activity patterns in radio-tracked mongooses. J. Wildl. Manage. 55(4):698-700.

 

36. Risenhoover, K. L. 1986. Winter activity patterns of moose in interior Alaska. Journal of Widlife Management. 50:727-734.

 

37. Rogowitz, G. L. 1997. Locomotor and foraging activity of the white-tailed jackrabbit (Lepus townsendii). Journal of Mammology. 78(4):1172-1181.

signal change (but called motion sensitive) (ampl/pitch? unclear),  presence/absence,

 

38. Scheibe, K. M., Eichorn, K., Schleusner, Th., Berger, A., and J. Langbein. 1995. Biorhythmic analysis of behavior of free ranging domestic and wild animals by means of a new storage-telemetry system. Pages 270-276 in C. Cristalli,  C. J. Amlaner, Jr., and M. R. Neuman,  eds. Biotelemetry XIII. Proceedings of the 13th International Symposium on Biotelemetry. March 26-31, 1995. Williamsburg, Virginia, U.S.A.

 

39. Shaefer, J. A. and S. N. Luttich. 1998. Movements and Activity of Caribou, Rangifer tarandus caribou, of the Tongat Mountains, Northern Labrador and Quebec. The Canadian Field Naturalist. 112(3):486-490.

 

40. Shields, L. J. 1980. The determination of free ranging rodent activity by telemetry. Pages 667-672 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

41. Singer, F. J., Otto, D. K., Tipton, A. R., and C. P. Hable. 1981. Home ranges, movements, and habitat use of European wild boar in Tennessee. J. Wildl. Manage. 45(2): 343-353.

 

42. Sun, Y. 1997. Tawny fish owl activity pattern. Wilson Bull. 109(4)737-741.

switch/signal change ampl, presence/absence, aural, + 

 

43. Sunquist, M. E. and G. G. Montgomery.  1973a. Activity pattern of a translocated silky anteater (Cyclopes didactylus). Journal of Mammalogy. 53(3), p 782.

 

44. Sunquist, M. E. and G. G. Montgomery.  1973b. Activity patterns and rates of movement of two-toed and three-toed sloths (Choloepus Hoffmanni and Bradypus infuscatus). Journal of Mammalogy. 54(4), pp.946-954..

 

45. Swanson, G. A., Kuechle, V. B., and A. B. Sargeant.  1976. A telemetry technique for monitoring diel waterfowl activity. J. Wildlife Mgmt. 40(1), pp. 187-190.

tip-switch, presence/absense,  strip-chart, visual, +

 

46. Taylor, K. D. 1978.Range of movement and activity of common rats (Rattus norvegicus) on agricultural land. Journal of Applied Ecology, 15 663-677.

pitch, presence/absence, aural / tape recorded / aural, +

 

47. Tester, J. R. 1987. Changes in daily activity rhythms of some free-ranging animals in Minnesota. Canadian Field Naturalist. 101:13-21.

 

48. Trivelpiece, W. Z., Bengtsom, J. L., Trivelpiece, S. G., and N. J. Volkman. 1986. Foraging behavoir of Gentoo and Chinstrap Penguins as determined by new radiotelemetry techniques.  Auk 103:777-781.

 

49. Van Ballenberghe, V. and D. G. Miquelle. 1990. Activity of moose during spring and summer in interior Alaska. Journal of Wildlife Management. 54:391-396.

 

50. Verts, B. J. 1963.Equipment and techniques for radio-tracking striped skunks. J. Wildl. Manage. 27(3):325-339.

 

51. Withgott, J. H. and C. J. Amlaner. 1995. Monitoring body temperature and nocturnal activity of snakes with implanted transmitters. Pages 257-262 in C. Cristalli,  C. J.

Amlaner, Jr., and M. R. Neuman,  eds. Biotelemetry XIII. Proceedings of the 13th International Symposium on Biotelemetry. March 26-31, 1995. Williamsburg, Virginia, U.S.A.  in Biotelemetry XIII.

 

52. Witt, H. 1980. Methods and experience gained from biotelemetry investigations of Mouflon (Ovis ammon musimon Schreber 1782). Pages 741-742 in C. J. Amlaner, Jr. and D. W. Macdonald, eds. A handbook on biotelemetry and radio tracking. Pergamon Press, Oxford, U.K.

 

53. Young, M. K. 1999. Summer diel activity and movement of adult brown trout in high-elevation streams in Wyoming, U.S.A. Journal of Fish Biology. 54:181-189.

 

54. Altmann, S.A. & Wagner, S.S. (1970) Estimating rates of behavior from Hansen frequencies. Primates 11, 181-183.

 

55. Chi-Tsong, C. 1979. One-dimensional digital signal processing. Marcel Dekker, Inc. New York.

 

56. Larkin, R.P., Raim, A., & Diehl, R.H. (1996) Performance of a non-rotating direction-finder for automatic radio tracking. Journal of Field Ornithology 67, 59-71.

 

57. Cochran, W. W., G. W. Swenson, Jr., Larry L. Pater. Radio Direction Finding for Wildlife Research",  Proc. 2001 Antenna Applications Symposium, Allerton Park  P 295-307 , ECE Dept., U.Mass. Amherst.

 

                                    Bibliography

 

Altmann, J. (1974) Observational study of behavior - sampling methods. Behaviour 49, 227-267.

 

Altmann, S.A. & Altmann, J. (2003) The transformation of behaviour field studies. Animal Behaviour 65, 413-423.

 

Atramentowicz, M. 1982. Influence de milieu sur l'activite locomotrice et la reproduction de Caluromys philander (L.). Review d'Ecolocgie. 36:373-395.

 

Beier, P and D. R. McCullough. 1990. Factors influencing white-tailed deer activity patterns and habitat use. Wildlife Monographs 109.

 

Bevins, J.S., Franzmann, A.W. & Schwartz, C.C. (1988) Determining moose activity budgets using leg-mounted tip-switch transmitters and a computerized data acquisition system. Alces 24, 22-33.

 

Cederlund, G., Bergström, R. and F. Sandegren. 1988. Winter activity patterns of females in two moose populations. Canadian Journal of Zoology. 67:1516-1522.

 

Cohen, A. & Denenber, V.H. (1973) Small-animal activity-analyzing system for behavioral studies. Medical & Biological Engineering 11, 490-499.

 

Cooper, J. A. and A. D. Afton. 1981. A multiple sensor system for monitoring avian nesting behavior. Wilson Bull. 93(3):325-333.

 

Gilmer, D. S., Kuechle, V. B., and I. J. Ball. 1971. A device for monitoring radio-marked animals. J. Wildl. Manage. 43: 1001-1007.

 

Gwinner, E. & Brandstatter, R. (2001) Complex bird clocks. Philosophical Transactions of the Royal Society of London Series B-Biological Sciences 356, 1801-1810.

 

Hansen, E.W. (1966) The development of maternal and infant behavior in the resus monkey. Behaviour 27, 107-149.

 

Heal, J.W. (1975) Animal activity monitor using a microwave doppler system. Medical & Biological Engineering 13, 317-317.

 

Jorde, D. G. and R. B. Owen. 1988. The need for nocturnal activity and energy budgets of waterfowl. Pp. 169-180 in Waterfowl in winter (M. W. Weller, ed.). Univ. Minnesota Press, Minneapolis, Minnesota.

 

Lawson, K., Kanwisher, J. and T. C. Williams. 1976. A UHF radiotelemetry system for wild animals. Journal of Wildlife Management. 40:360-362.

 

Merrill, S.B. & Mech, L.D. (2003) The usefulness of GPS telemetry to study wolf circadian and social activity. Wildlife Society Bulletin 31, 947-960.

 

Pittendrigh, C.S. (1993) Temporal Organization - reflections of a Darwinian clock-watcher. Annual Review of Physiology 55, 16-54.

 

Sargeant. A. B., Forbes, J. E., and D. W. Warner. 1965. Accuracy of data obtained through the Cedar Creek automatic radio tracking system. Minnesota Museum of Natural History Technical report #10. December 1965. pnm    

 

Theuerkauf, J. and W. Jedrzejewski. (2002)  Accuracy of radiotelemetry to estimate wolf activity and locations.  Journal of Wildlife Management 66(3): 859-864.  

 

Van Oort, B.E.H., Tyler, N.J.C., Storeheier, P.V. & Stokkan, K.A. (2004) The performance and validation of a data logger for long-term determination of activity in free-ranging reindeer, Rangifer tarandus L. Applied Animal Behaviour Science  89, 299-308.

 

Verner, J. 1965. Time budget of male long-billed marsh wren during the breeding season. The Condor. Volume 67 pp. 125-139.

 

Young, M. K., Wilkison, R. A., Phelps III, J. M., and J. S. Griffith. 1997. Contrasting movement and activity of large brown trout and rainbow trout in Silver Creek, Idaho. Great Basin Naturalist. 57(3):238-244.

Appendix I: ARU technical
Appendix II: Worklists, Number of Channels
Appendix III: Additional figures
Appendix IV: Viewing and analysis tools
Appendix V: Review of methods literature

Copyright © 2006
Permission to copy any material in this paper is hereby granted on the condition that any
publication of copied material is acknowledged to the author, title, and website:
W. W. Cochran. 2006. Measuring animal activity from signal strength recordings. www.sparrowsystems.biz