Survey design: transect close together, but relatively low speed platform

54 views
Skip to first unread message

Morgane Pommier

unread,
May 1, 2024, 6:44:33 AMMay 1
to distance-sampling
Hi, 

I am currently designing a distance sampling survey for seabirds and marine mammals using aerial imagery from a drone.
Given the strip width, I need to space the transects quite close together (1km ?) to achieve our target coverage and the entire study area is probably not going to be covered in one day given the platform speed (maybe the same as a boat ?).

I am conscious about double counting and the effect it could have on abundance estimates. 
Is there a way to statistically account for this, or a strategy to adopt in the design itself to chose which line to cover in a day (maybe jump lines to cover the whole area in one day, but come back to other lines the next day or something like this) ? Or should I stratify the area ? 

Thank you in advance for any advice, 

Morgane

Eric Rexstad

unread,
May 1, 2024, 7:03:35 AMMay 1
to Morgane Pommier, distance-sampling
Morgane

Thanks for your questions and your diligence to think about these issues at the time of design.

I am curious how you arrived at a value for "target coverage". When determining effort needed to achieve a desired level of precision in conventional distance sampling, effort is measured in kilometers of effort, not proportion of study area sampled. See a discussion of this in Section 2.4 of Buckland et al. (2015). It is critical that the sample is representative of the study area, this can be achieved without covering a large proportion of the study area (think about pollsters gauging public opinion; they don't ask 25% or even 5% of the population for their opinions, yet by asking a group representative of the population, polls are often quite accurate).

Your second paragraph touches upon advice we often give. If transects cannot all be visited in a day, then plan your flights such that they fly alternate (or every third, or every fourth) transect. Then on subsequent days, begin to fill in the missed transects. This way, even if the survey ends after the first day (because of weather or drone malfunction), you still have a systematic survey that extends from one edge of the study area to the other and is therefore representative even in the presence of a possible gradient.

Stratification could be used for logistic reasons, if perhaps, drones operators need to reposition. But the alternate transect approach is likely to be sufficient.
From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Morgane Pommier <pommier....@gmail.com>
Sent: 01 May 2024 11:44
To: distance-sampling <distance...@googlegroups.com>
Subject: [distance-sampling] Survey design: transect close together, but relatively low speed platform
 
--
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 view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/66f66232-392f-42e1-96f5-47aae6e63ef9n%40googlegroups.com.

Res. Ass. Uğur ÖZSANDIKÇI

unread,
May 2, 2024, 7:32:04 AMMay 2
to distance-sampling
Dear Eric, Morgane and all,

I have been thinking about similar works with drones or aerial imagery for some time, and when I came across it here, I wanted to ask about it. 
The basic principle of the distance sampling method is that as the distance between the object and the observer increases, the probability of detecting the object decreases. Therefore, the probability of detection and then the effective strip width is estimated from the perpendicular distances of the detected objects to the observer. The question that confuses me is: does this principle still hold true for aerial imagery? Does the probability of detecting an object at a pixel on the transect and the probability of detecting an object at a pixel further away from the transect change with distance? My other question is about the edge-to-edge suggestion. Although this method can be applied in some aerial surveys, it may not be possible to apply it in large areas in ship-based marine surveys. In large-scale ship surveys, it sometimes takes more than 20 days to finish sampling the transects. Is this a factor that reduces the strength of the design? I wanted to ask you since I have received many questions on this subject.

Best regards, 

-- Uğur
1 Mayıs 2024 Çarşamba tarihinde saat 14:03:35 UTC+3 itibarıyla Eric Rexstad şunları yazdı:

Morgane Pommier

unread,
May 2, 2024, 8:44:20 AMMay 2
to distance-sampling
Hi  Uğur, Eric, 

Thank you both for your replies.

Uğur, I don't have an answer to your second question, but I may have some ideas regarding the first one:
I think (from some reading around, but Eric or someone else can correct if I'm wrong) that if you use a digital approach to aerial surveys (cameras), then you can assume that all animals in the pictures are detected and that the detection probability is uniform and equal to 1, or at least independent from the distance to the transect line. Then you don't need to fit a detection function to your data and can derive the density estimate directly using a strip transect approach: 
Density =  (Total number of animals * area of entire study region)/(area surveyed, or in other words strip width * total line length). 

Hope this can help, 
Kind regards,

Morgane

Stephen Buckland

unread,
May 2, 2024, 9:01:24 AMMay 2
to Morgane Pommier, distance-sampling
Yes Morgane, that is correct. But it may be that some animals are in the strip but unavailable (e.g. diving animals), I. Which case you need a way to estimate the available proportion. Also, if you use an automated way to identify objects in the images, you may need a way to estimate a false positive rate - and perhaps a false negative rate, though that might be addressed by estimating availability.

Steve 

Stephen T. Buckland
CREEM, The Observatory, Buchanan Gdns, St Andrews KY16 9LZ, Scotland

The University of St Andrews is a charity registered in Scotland:No SC013532
Sent: Thursday, May 2, 2024 1:44:19 PM
To: distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] Survey design: transect close together, but relatively low speed platform
 

Tiago Marques

unread,
May 2, 2024, 9:04:57 AMMay 2
to Morgane Pommier, distance-sampling
Hi folks,

It really boils down to what is fair to assume. If you have a down facing camera that covers perfectly a field of view, maybe you can assume that you do not need to model detectability (you might need to model availability, e.g. submerged animals) as a function of distance. But if you have a camera facing sideways it might be that objects further out are harder to detect, so it is up to you to decide what your settings allow you to assume to be reasonable.

Re the edge-to-edge, most large scale surveys can't do it, so that is just something you have to be aware. It means that often the survey effort is actually considerably different from the planned effort. You have to live with it. A possible way to try to see if that was impactful in the results is to compare a design based with a model based approach. If there are big differences, maybe the design oversampled some areas at the expense of undersampling others, leading to the differences. It will be up to you to decide if you prefer the model based or the design based estimates.

Cheers,

T
Sent: 02 May 2024 13:44
To: distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] Survey design: transect close together, but relatively low speed platform
 

Morgane Pommier

unread,
May 2, 2024, 9:06:55 AMMay 2
to distance-sampling
Yes true, the availability bias or imperfect detection issues are still there...
Thank you for the precision !

Morgane

Morgane Pommier

unread,
May 2, 2024, 9:10:06 AMMay 2
to distance-sampling
Just reading your answer now Tiago, thanks as well for your inputs ! I hadn't considered the scenario of an angled camera but I see how it could impact the detections indeed.

Stephen Buckland

unread,
May 2, 2024, 9:30:28 AMMay 2
to Tiago Marques, Morgane Pommier, distance-sampling

Adding to Tiago’s comments, usually, a camera at low altitude pointing downwards covers a very narrow strip relative to observers on a plane.  However, use of high-resolution imagery means that you can operate at much higher altitude than would typically be used for observers on a plane.  That widens the strip, and also means that the distance to an animal at the outer edge of the strip is almost the same as for an animal immediately below the camera.  I’m not familiar with camera surveys where the camera is appreciably angled away from down, but if this is done, then detectability is likely to drop with distance from the line, as Tiago suggests.

 

Steve

Res. Ass. Uğur ÖZSANDIKÇI

unread,
May 2, 2024, 6:23:39 PMMay 2
to distance-sampling
Thank  you all for your clear and quick responses.

-- Uğur

2 Mayıs 2024 Perşembe tarihinde saat 16:30:28 UTC+3 itibarıyla Stephen Buckland şunları yazdı:

Marina Costa

unread,
May 3, 2024, 5:55:29 AMMay 3
to distance...@googlegroups.com
Hi all,

about the camera frame, in the study we are running with a medium-size drone (7m wingspan) we face the issue described by Tiago. Our camera faces sideways to avoid the glare and the animal/pixel size changes with the perpendicular distance to the track-line. This might be not an issue for large animals (i.e., fin whale in the attachment) but surely affects dolphins. I attached a picture showing how the sightability decreases with the distance. From pictures on site, the first and third individual in the attachment are very likely the same.

We definitely would like to apply distance sampling to drone imagery, provided we will have a sufficient number of sightings.

A huge hug,
Marina
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/AS8P251MB093503E1700CD0324DA36EF6AA182%40AS8P251MB0935.EURP251.PROD.OUTLOOK.COM.

--
 
logo

Marina Costa, Ph.D.

Member of the Board of Directors

Tethys Research Institute
www.tethys.org

 

E: marinz...@gmail.com T: +39 328 7861 713

 
logo logo logo

 

This e-mail transmission is intended only for the addressee (s) indicated above. It may contain information that is privileged, confidential or otherwise protected from disclosure. Any review, dissemination or use of this transmission or its contents by persons other than the addressee is strictly prohibited. If you have received this message in error, please be so kind to contact us by e-mail at the mail (tet...@tethys.org) and delete the message from your server.
Screenshot 2024-05-03 114728.png

Eric Rexstad

unread,
May 3, 2024, 6:06:08 AMMay 3
to Marina Costa, distance...@googlegroups.com
Thanks for sharing the images from your drone Marina. Helpful to see the data derived from different platforms.
Reply all
Reply to author
Forward
0 new messages