Butterfly Distance Sampling Field Design

[Michelle P]

Hi everyone, I am designing distance sampling for a species of small skipper in Ontario, which exists at low abundances/densities. At one site we will be able to use transects, and a separate site we will use points. I have some questions regarding field methods and survey design, especially as they relate to butterfly surveys:

1. I want to confirm that the net-sweeping method (mentioned by A. Ellingson: here) is appropriate. I have hesitations as it seems like it would increase the quantity of detections near the observer (g(0)), while butterflies outside of the sweeping radius would remain undisturbed, which could affect the detection function. Is this sweeping methodology appropriate for distance sampling transects, AND/OR distance sampling points?

2.  I've read on this forum that transects as short as 20m can be appropriate for butterflies. For the site where I will be able to use transects, I have created a Systematic Segmented Trackline Sampling Design in Distance. My design consists of 20m long segments separated from one another (end of one to start of next) by 1m. They are spaced 25m apart, and the design produces approximately 160 segments distributed across my site, which I intend to randomize and survey. My reason for creating short segments is that I plan to eliminate segments that are impassable due to dense shrubs, via ground truthing. Because my species is very closely associated with patches of their host plant (a low-growing plant), I think that eliminating dense-shrub transects that do not constitute habitat of the species is defensible, and that I'll be able to analyse my data accurately by also eliminating the dense shrub areas from the total "habitat area" when creating abundance estimates. 

Can anyone confirm that sweeping a net, using short segments (20m) and eliminating those that are impassible are reasonable field design choices?

This forum has been immensely helpful to me so far, thank you all. 

Michelle

[Michelle P]

3. I plan to use double-observer methods for both sites, (at transects and points). From what I have read, this should work out logistically in the field, but there may be very little variation in detections made by each observer (they may detect all the same animals), since we are working at such a small spatial scale. I don't anticipate that this will be an issue during analysis, or foresee any other survey protocol related issues, but if anyone knows differently I would appreciate any advice. 

[Stephen Buckland]

Michelle, a few comments. For most butterfly species, point transects are not a good idea. If you stand at a point for any length of time, butterflies will fly past eventually. But distance sampling is a snapshot method - you want to record the distance of butterflies from the point at a single moment in time. If you wait for them to fly by, you will on average record a distance that is too small, and you will overestimate abundance. If these skippers seldom fly, the method may work, but sample sizes are likely to be very small. I’m guessing that this species can be difficult to spot when resting. Sweep netting to increase g(0) is fine, unless you otherwise miss many near the line. Then sweep netting may generate a very spiked detection function that you may have difficulty modelling, and high uncertainty in any abundance estimate. But in that case, without sweeping, your g(0) would be a lot less than one, and without sweeping, you’ll underestimate abundance. I guess that is why you plan double observers, but you're right, they will both tend to see the more visible butterflies, so you will still underestimate. If you don't get many detections beyond the reach of the net, I would truncate detections close to the maximum distance you reach. A long-handled sweep net might help. Flying butterflies will cause upward bias. You can either only record resting butterflies and separately estimate proportion of butterflies that are flying at any one time or adopt a procedure to reduce bias, such as record their distance as they pass 'abeam' of you, and not record ones that don't pass abeam while in sight. I don't see any reason for having line segments with just 1m separation as you won't be able to assume that data from such segments are independent. I would just have long lines across your study area. Perfectly ok to omit sections of line through unsuitable habitat if you expect few or no butterflies in those sections, and yes, you would subtract such areas from the size of your site.

I don't think this will answer all your questions, but hopefully it will help you to work out a suitable scheme!

Steve Buckland 

Stephen T. Buckland

CREEM, The Observatory, Buchanan Gdns, St Andrews KY16 9LZ, Scotland

e-mail st...@st-andrews.ac.uk

The University of St Andrews is a charity registered in Scotland:No SC013532

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Michelle P <michelle...@gmail.com>
Sent: Thursday, May 4, 2023 8:34:34 PM
To: distance-sampling <distance...@googlegroups.com>
Subject: [distance-sampling] Re: Butterfly Distance Sampling Field Design

 

--
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/66cc5f6d-a1a9-4f9f-bd95-0696c0eaf751n%40googlegroups.com.

[Len Thomas]

Just to augment what Steve said regarding unsuitable habitat (just
focussing on that one issue):

As Steve said, it's fine to omit sampling the parts of your transect
lines that go through unsuitable habitat, where you do not expect any
butterflies. Then, when it comes to the analysis, you have a few
options, all of which should lead to the same estimates of abundance.

The first is to tell the software you surveyed the entire lines, and
that the size of the study area is the whole area within which you
placed the lines (both suitable and unsuitable habitat). Assuming you
laid your lines out at random (systematic random is fine), this will
yield an estimate of the average density of butterflies in the whole
area (both suitable and unsuitable habitat). Multiplying density by
study area size gives an abundance of butterflies in the whole area.

The second is to tell the software you only surveyed the parts of the
lines in suitable habitat. For example let's say that the entire lines
are of length L but that length L_u was found to be unsuitable, leaving
L_s suitable. So for survey effort, you'd enter just the suitable line
length per transect, which would total L_s. Then you get a density of
butterflies in *suitable* habitat. So, to get the correct abundance,
you need to tell the software what the area is of suitable habitat in
your study area. You may not know this, but assuming the proportion of
suitable habitat on your transect lines is an unbiased estimate of the
proportion of suitable habitat in the study area, then you can take the
study area size and multiply that by L_s / L. As I expect you can see,
you've reduced the line length by L_s / L and then reduced the study
area size by L_s / L, so the two balance out and you get the same
estimate of abundance as the first method above.

One difference, however, is that the variance estimate will be
different. As you know, we use the line-to-line variation in detection
rate as part of variance estimation, and in the first case part of the
variation in detection rate will be due to variation in the proportion
of unsuitable habitat. So likely the first method will give you a
higher variance than the second. On the other hand, using the second
method, we have estimated the proportion of unsuitable habitat in the
study area but not accounted for uncertainty in this estimation process
in our overall variance. It would be possible to account for this, by
taking the proportion of unsuitable habitat by line, then taking a (line
length weighted) mean and a weighted variance, and entering these as
multipliers. But that seems a hassle to me, so overall I favour the
first of the two approaches I outline above.

However, if you know what proportion of your study area is suitable
habitat (e.g., from aerial photos) then the second method could be used
but with a true value for the suitable area size within the study area
rather than an estimate.

Feel free to pitch in, anyone, if you disagree with the above!

Cheers to all, Len

--
I am a member of the University and College Union and am currently
participating in industrial action to improve UK higher education
staff pay, equality and working conditions. For more
information, please see
https://www.ucu.org.uk/article/12469/FAQs#Pay_and_working_conditions_dispute

--
Len Thomas (he/him) len.t...@st-andrews.ac.uk lenthomas.org @len_thom
Centre for Research into Ecological and Environmental Modelling
and School of Mathematics and Statistics
The Observatory, University of St Andrews, Scotland KY16 9LZ
Office: UK+1334-461801 Admin: UK+1334-461842

While I may be sending this email outside of my normal office hours,
I have no expectation to receive a reply outside of yours.

The University of St Andrews is a charity

registered in Scotland, No SC013532.

[Michelle P]

Thank you Stephen and Len! So much appreciated. I'll keep all of these things in mind as I design my surveys.