how to analyze repeated point counts in Distance

[ca...@kauaiforestbirds.org]

hi all, I've seen several related posts but so far none that answer my question.

We have made three visits over the course of several weeks to each of 30 point count stations, where we performed an 8-minute count on each visit.

We have imported all of the data into Distance, accounting for the three repeat visits as "survey effort" at the "point transect" level. So for each point, we have the name of the point, and the survey effort = 3. Then at the sample level, we have each observation with its distance, as well as covariates relating to the time and weather of each visit, and the visit number (1-3).

My question is: where in the modeling phase (the "inputs") do we bring the survey effort in? i.e., where do we tell Distance that we have repeated counts at each station, as opposed to all of those observations coming from a single visit.

I understand that the point may be moot because we have equal numbers of visits at all stations....but even if so, for the future, when that is likely to not be the case, I'd like to know how to model survey effort.

Thanks!
Cali

[Eric Rexstad]

Cali

Not sure I understand your question.  Sounds like you've set up your project properly.  You have recorded number of visits in the "Survey effort" field of the transect layer along with covariates (including visit number) at the observation layer.

Distance calculates encounter rate as the number of detections per unit of effort (visit).  Consequently by specifying 3 visits to the point, Distance calculates the proper encounter rate.  There is no need for the number of visits to be equal for all points.  Perhaps on one visit, you were unable to get to some of the points, therefore the number of detections on those points will be a function of 2 visits rather than 3; you merely indicate "Survey effort" of 2 for those points.

Your analysis will not model survey effort (it is known and not estimated), but the number of detections is a function of effort; that is taken into account in the manner I described.

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampl...@googlegroups.com.
To post to this group, send email to distance...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/bbf2e5ca-ffd9-4339-b4b9-26b591a99896%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

-- 
Eric Rexstad
Research Unit for Wildlife Population Assessment
Centre for Research into Ecological and Environmental Modelling
University of St. Andrews
St. Andrews Scotland KY16 9LZ
+44 (0)1334 461833
The University of St Andrews is a charity registered in Scotland : No SC013532

[Lisa Cali Crampton]

hi Eric, thanks for the quick replay.

So if I understand what you're saying, simply by specifying survey effort in the data import stage, I've done all I need to do. I don't need to refer to that column indicating survey effort in any of the further model input windows. Is that right? Great!

Thanks

Cali

--

Lisa "Cali" Crampton
Project Leader
Kauai Forest Bird Recovery Project
PO Box 27 (USPS mail) or 3751 Hanapepe Rd (courier packages)
Hanapepe HI 96716

808.335.5078

[Eric Rexstad]

Exactly right Cali.  The "Survey effort" field is a mandatory field (visits for points, line length for transects).  Distance MUST have that information because it is "hard wired" into Distance's calculations.  Therefore, you need not explicitly "point" to that field, Distance looks for it automatically.

[Rebecca]

Hi there,

I believe I am having a similar issue. I am carrying out a point transect survey with 4 points, each visited 80 times. I put my effort for each observation as 80 but the covered area has been calculated by multiplying the area of each point by 80 - I assume this is because it is assuming I have sampled 80 points per area (4 areas) whereas each area is only 1 point and repeats were at the same location meaning the covered area is 80x too large.

Hope you can help!

Thank you,

Rebecca 

[Rebecca]

To add another question for the same R script (above was talking about input file for ds), for my dht2 output the 'Area' is coming out as the individual area for each region whereas the 'CoveredArea' is presented as the total, meaning my abundance estimates are lower than the actual sample since the covered area is greater than the total.

[Eric Rexstad]

Rebecca

Not to worry. Think about "CoveredArea" in this way: it is the area sampled taking into account multiple visits to the transect.

Take this fictious example: I visit a square meter of ground and count 11 beetles in that square meter. Tomorrow I visit the same square meter again and count 9 beetles. If detectability is pefect, what is an estimate of beetle density from that survey? It is the total number of beetles (20) divided by the area sampled. I would argue that the area sampled is 2m^2, making the density estimate 10 beetles per m^2.

We can follow these calculations with a point transect survey with multiple visits. Use the data set "wren_snapshot" that ships with the Distance package. It represents an analysis published in

• Buckland, S. T. (2006). Point transect surveys for songbirds: Robust methodologies. The Auk, 123(2), 345–357. https://doi.org/10.1093/auk/123.2.345

Thirty two point transects were visited two times and the data were truncated at 110m. One hundred seventeen (117) detections of winter wrens were made by Prof Buckland.

The result of a fitted hazard rate model is summarised below:

The density estimate (1.023) is the same as published in Buckland (2006):

How was that estimate derived:

Estimated birds in the covered area = birds detected / detection probability = 117/0.47 = 248.9

Estimated density = estimated birds in covered area / covered area

Covered area = number points * number visits * pi * (truncation distance)^2 / (meters per hectare) # unit conversion

Covered area = 32 points * 2 visits * 3.14159 * (110m)^2 / 10000

Covered area = 243.28 hectares # note this covered area is much larger than the size of the study area

Estimated density = 248.9 birds / 243.28 hectares = 1.02 birds per hectare

Does this relieve your anxiety?

An aside: having 4 replicate point transects (spatial replicates) visited 80 times is not going to produce terribly robust estimates. Temporal replication (repeated visits) is not a substitute for spatial replication. Encounter rate variance (one component of uncertainty in density estimation) is computed on the basis of variability between spatial replicates. Having a small number of spatial replicates leads to poor estimates of this encounter rate variance. A better survey design to employ in the future is to have more spatial replicates visited fewer times, resulting in more reliable estimates with effectively the same sampling effort.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Rebecca <rmaca...@gmail.com>
Sent: 30 September 2024 20:35
To: distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] how to analyze repeated point counts in Distance

 

To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/f126a544-fece-4699-9456-9224ff999355n%40googlegroups.com.

[Rebecca MacArthur]

Hi there,

That was very helpful thank you so much for your response. Would I therefore be correct in saying that the effort is the number of points per area x number of visits to the points (does the amount of time surveying at each point come in anywhere?).

Thanks again,

Rebecca 

[Eric Rexstad]

Rebecca

Duration of a point count visit (5min, 10min, ...) is only relevant if performing avian cue counting surveys where a detection is of a "song" rather than an animal. Distance sampling can then be used to estimate "song" density in units of songs/area/time. Then a "song production rate" multiplier (measured in units of songs per time) is applied to convert song density into bird density.

---

From: Rebecca MacArthur <rmaca...@gmail.com>
Sent: 02 October 2024 01:35
To: Eric Rexstad <Eric.R...@st-andrews.ac.uk>
Cc: distance-sampling <distance...@googlegroups.com>

[Stephen Buckland]

Adding to Eric’s comment, the method he describes is the cue-count method.  In standard point transect sampling, conceptually, recording at a point is a ‘snapshot’ – that is, instantaneous.  The method gives upwardly-biased density estimates if birds are moving around during the period spent recording at a point.  For a discussion of this issue and how to avoid this bias, see

Buckland, S.T. 2006.  Point transect surveys for songbirds:  robust methodologies.  The Auk 123, 345-357.

 

Steve Buckland

To add another question for the same R script (above was talking about input file for ds), for my dht2 output the 'Area' is coming out as the individual area for each region whereas the 'CoveredArea' is presented as the total, meaning my abundance estimates are lower than the actual sample since the covered area is greater than the total.

To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/DBAPR06MB6694226BB8E5CF0D7DE26CF0EA702%40DBAPR06MB6694.eurprd06.prod.outlook.com.

[Rebecca MacArthur]

Thank you both for your help. I am indeed looking at recording bird song (hence why sample is limited by number of recording devices) but will take each song as an individual by cutting repeat species calls within the same 5 minute period. Since each call will thus be assumed to represent an individual would it be okay to maintain the original method rather than the cue counting method? Additionally, for repeats of an individual should the first call or the average call distance be taken?

Kind regards, Rebecca

On 2 Oct 2024, at 02:52, Stephen Buckland <st...@st-andrews.ac.uk> wrote:



<image001.png>

 

The density estimate (1.023) is the same as published in Buckland (2006):

 

<image002.png>

 

How was that estimate derived:

 

Estimated birds in the covered area = birds detected / detection probability = 117/0.47 = 248.9

 

Estimated density = estimated birds in covered area / covered area

 

Covered area = number points * number visits * pi * (truncation distance)^2 / (meters per hectare) # unit conversion

 

Covered area = 32 points * 2 visits * 3.14159 * (110m)^2 / 10000

Covered area = 243.28 hectares # note this covered area is much larger than the size of the study area

 

Estimated density = 248.9 birds / 243.28 hectares = 1.02 birds per hectare

 

Does this relieve your anxiety?

 

An aside: having 4 replicate point transects (spatial replicates) visited 80 times is not going to produce terribly robust estimates. Temporal replication (repeated visits) is not a substitute for spatial replication. Encounter rate variance (one component of uncertainty in density estimation) is computed on the basis of variability between spatial replicates. Having a small number of spatial replicates leads to poor estimates of this encounter rate variance. A better survey design to employ in the future is to have more spatial replicates visited fewer times, resulting in more reliable estimates with effectively the same sampling effort.

 

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Rebecca <rmaca...@gmail.com>
Sent: 30 September 2024 20:35
To: distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] how to analyze repeated point counts in Distance

 

To add another question for the same R script (above was talking about input file for ds), for my dht2 output the 'Area' is coming out as the individual area for each region whereas the 'CoveredArea' is presented as the total, meaning my abundance estimates are lower than the actual sample since the covered area is greater than the total.

<image003.png>

[Rebecca MacArthur]

If I must use the cue-count method, I am unsure as to how this could be applied since I am looking at multiple species. I have read the winter wren point count analysis but for a multi-species model I am unsure as to how the data could be laid out since I assume the cue rate would apply to the each species average rather than an individual, but distances to the sound are surely to an individual rather than species average. Thank you again for your advice, I really appreciate it.

Rebecca 

[Eric Rexstad]

Rebecca

I don't understand what you mean about "cutting repeated species calls". If you are taking the cue counting approach (which you must be doing if using autonomous recorders rather than human observers), you record each call as a  detection and you measure distances (you don't describe how you are doing this) for each and every call.  Cue counting makes no assumptions about calls belonging to individuals.

Regarding your next email, you need to measure cue production rate for a number of individuals (because all individuals do not produce cues at the same rate). You then need to measure the variability in call rates between individuals to incorporate variability in call production rate at the population level as part of uncertainty in your estimates of density. Have a look at one of the distance sampling text books to understand all the components that go into estimating animal density from cue counts.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Rebecca MacArthur <rmaca...@gmail.com>
Sent: 02 October 2024 14:39
To: Stephen Buckland <st...@st-andrews.ac.uk>
Cc: Eric Rexstad <Eric.R...@st-andrews.ac.uk>; distance-sampling <distance...@googlegroups.com>

To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/58661ABB-15EE-4523-A4E3-A69CD776F875%40gmail.com.

[Rebecca MacArthur]

Thank you Eric, would I therefore be correct in saying that if the same species was heard making 5 calls in a 5-minute period it would be taken as 1 individual with a cue rate of 1 and this is repeated for every individual and an average and SE taken for each species? The input data would therefore have a list of every individual with the average cue rate for the species and the distance from the recorder.

[Stephen Buckland]

You would input the cue rate as the average (across the monitored individuals) and standard error, rather than enter each monitored individual.  For the individuals that you include in your cue rate estimation, you need to be able to detect all songbursts (or calls) within the time that you monitor them.  Try reading up on the method, either in the 2006 Auk paper, or the 2015 distance sampling book.  The book covers more issues, so would be the better option.  I’ll send you off-list the Auk paper.

 

Steve Buckland

[Rebecca MacArthur]

Thank you so much. That makes sense and has really helped my data get going! Just one final hurdle I am trying to deal with - the cue rate. I was wondering if for the above paper (

https://academic.oup.com/auk/article/123/2/345/5562622

) you could give me an idea of the cue rates per min / hour that you found for each species, respectively to give me an idea of whether I'm on the right track. I have been trying to assess how many cues per hour at multiple sites/times of day but am slightly confused as to whether I should be recording it as the number of times a bird is vocal for a significant period of time or the number of separate calls a bird makes e.g. for a crow would it be the number of times it produces a series of calls, or the number of individual calls?

Hope this makes sense and thank you again so much for your help.

Kind regards,

Rebecca 

On Wed, Oct 2, 2024 at 11:58 AM Stephen Buckland <st...@st-andrews.ac.uk> wrote:

Auk paper attached.

[Eric Rexstad]

Rebecca

The cue rates (and their standard errors) for all four species are found in Table 1 of Buckland (2006)

---

From: Rebecca MacArthur <rmaca...@gmail.com>
Sent: 10 October 2024 02:03
To: Stephen Buckland <st...@st-andrews.ac.uk>; Eric Rexstad <Eric.R...@st-andrews.ac.uk>; distance-sampling <distance...@googlegroups.com>

[Stephen Buckland]

Adding to that, how you define a cue depends on the species.  In my case, for three of the species, it was a ‘songburst’ – a single bout of song followed by a period of silence.  In the case of the European robin, its song can go on for an extended period, but it is made up of song ‘phrases’, so each phrase was taken as a cue.  That explains the higher cue rate for that species.  Carrion crows in the UK typically give 3 calls in quick succession, so if I was applying the method to them, I would take the set of 3 calls to be a single cue.

 

Steve Buckland

[Rebecca MacArthur]

Thank you, that is what I had thought but in this table the winter wren cue rate is noted as 7.28 whereas in the distance software analysis example it is noted as 1.4558 and the standard error also decreases. I was wondering why this changes as I thought it was the same example. 

Kind regards, Rebecca

On 10 Oct 2024, at 01:13, Eric Rexstad <Eric.R...@st-andrews.ac.uk> wrote:



Rebecca

The cue rates (and their standard errors) for all four species are found in Table 1 of Buckland (2006)

<image.png>

[Rebecca MacArthur]

Additionally, when determining cue rate, should I record for a full hour and then divide by 60 to get the cues per minute (meaning less commonly heard species would have a lower cue rate) or just count cues made whilst singing?

Many thanks again,

Rebecca 

[Eric Rexstad]

Rebecca

The cue rate value depends upon the duration over which the cue rate is measured. Prof Buckland conducted his cue counts for 5 minutes, hence the cue rate per 5 minutes is 7.28. Expressed as number of cues per minute, the value is 1.456.

---

From: Rebecca MacArthur <rmaca...@gmail.com>
Sent: 10 October 2024 15:10

[Stephen Buckland]

I believe that the rate in the paper is cues per 5mins, which was the time spent at each point.  The rate in the example will be the rate per minute.  Effort would be number of 5-min periods in the first case, and number of minutes in the second.  Both ways of analysing the data will give the same estimate.

 

You must record cue rate for a representative sample of birds.  Do not record just when they’re singing – it must be an average across birds.  If for example you’re estimating number of territories of a songbird, with an assumed 1 male per territory, you would need to select a representative sample of territories, and record for a pre-determined time period.  That doesn’t need to be a long time period.  Because different individuals may have different cue rates, monitoring a number of birds, each for a short period, is preferable to monitoring a few birds, each for a long period.

 

________________________________________________________________________________________________________________________________________________________________________________________________________________

 

Additionally, when determining cue rate, should I record for a full hour and then divide by 60 to get the cues per minute (meaning less commonly heard species would have a lower cue rate) or just count cues made whilst singing?

Many thanks again,

Rebecca 

 

 

From: Rebecca MacArthur <rmaca...@gmail.com>
Sent: 10 October 2024 15:10
To: Eric Rexstad <Eric.R...@st-andrews.ac.uk>
Cc: Stephen Buckland <st...@st-andrews.ac.uk>; distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] how to analyze repeated point counts in Distance

 

Thank you, that is what I had thought but in this table the winter wren cue rate is noted as 7.28 whereas in the distance software analysis example it is noted as 1.4558 and the standard error also decreases. I was wondering why this changes as I thought it was the same example. 

[Rebecca MacArthur]

Brilliant, I can't thank you enough for all of your help.

Kindest regards,

Rebecca