Thanks Morena and Lucy for your clarification on one of the terms i was having more trouble with, that makes more sense now. My question would be, if the ESW is thin, might you decide to do more transects to increase detectability?
I have been struggling to pick what has not been covered in the discussions, but I think I will go for the assumptions of distance sampling. I used these in my briefing with the team on Monday (we started counting yesterday).
Basically, emphasis on these assumptions to the team, helped with ensuring that the teams are focused and take accurate measurements. This ensures data reliability for modelling detection probabilities. In my talk, I used the âno movementâ assumption, (which states that animals do not move in the presence of the observer) - to encourage discipline in team members during counting.
1 Like
Re-reading parts of Buckland et al 2021 this week, I was reminded of their list of ways to measure distance (section 7.4, p254):
- Direct measurements, for surveys when detections are rare and close-ish to the line, e.g. using a tape measure - probably most useful for sign like scats/footprints
- GPS locations of observer and sighting, again when detections are rare & close-ish to the line so that going to the site of the observation doesnât take too long or disturb the species of interest - probably most useful for sign like nests
- Optical range-finders, which give you a distance when you manually focus on the object
- Laser range-finders, which measure the distance by bouncing a laser beam back to the range-finder
- Clinometers, which use trigonometry based on the eye-height of the observer and angle of declination to the object observed (presumably also requires you to estimate the size of the animal?)
- Binoculars with reticule marks - more commonly used on ship-board surveys. Again requires you to know the size of the object or have a line-of-sight on the horizon
- Estimation based on training
In situations where it is hard to estimate distances accurately, it is also possible to group distances into bands, and use those groups as our distance bins in analysis
One thing to avoid is rounding/heaping, where observers record the nearest round number e.g. 50m instead of 45 or 55. This can make fitting the detection function more difficult because you get artificial bumps in the pattern of detections at the rounded distances
I hope that helps!
Matt, I love this visual explanation! Thanks for taking the time to summarise your ideas in this way
During analysis:
- We test different-shaped detection functions - half-normal, uniform, hazard-rate
- R calculates the shape coefficients to get the best fit to our data, e.g. broad or narrow shoulder beside the transect, trailing off fast or slow, long or short tail etc
The pattern of our detections, and the size of our ESW, depends on vegetation, detection methods, animal behaviour etc. All weâre doing is getting a good match to that shape during analysis. The detection function doesnât need to influence our surveys, and neither does ESW
We might have many sightings within a tiny ESW (small reptiles or invertebrates along a linear feature), or many sightings within a large ESW (large ungulates in open savannah). Likewise, we may have very few sightings in either extreme of ESW. This doesnât affect the accuracy or precision of our estimates, it merely reflects the environment and species weâre working with
So the number of sightings we need, and distance we need to walk/sample size (number of transects) is independent of ESW
Does that makes sense?
1 Like
Does my answer to Matt above clarify this for you, Fergus?
In brief, we donât need to adjust the number of transects or try to increase detectability in response to the size of the ESW
If our number of detections (encounter rate) is too low to gain a reliable density estimate, we may need to re-survey the same transects or add new transects to gain enough data to model detectability effectively, but this isnât necessarily connected to ESW
Hi Lucy, yes this helps thank you for explaining!
1 Like
Detection is certain
From my understanding of this assumption it would seem that distance sampling is based on underlying knowledge around the habitat preferences of the species of interest. It is this knowledge that is used to select the study sight and randomly place the transects. In our case we have historic data on White backed vulture nesting sites and tagged birds we can use verify these areas. It is on these areas that we will place the transects and do the density estimates.
Yes, knowing the basic ecology of your target species is so important to help you design surveys to detect it/them effectively
The assumption Detection is certain is more specific than that you will detect your species of interest at some point during your surveys. Can you elaborate on what this assumption is, including why it matters?
The underlying assumption is that the species on or near the transect or point can be detected. Design of the survey should ensure that this is met. Hence my earlier submission that to ensure this certainty of detection is met the observation protocol should ensure g(0)= 1.
Some have used used video cameras and dogs to aid in counting observations near the line.
Care should be made as well so that the detection function does not fall steeply with distance from the line.
This assumption translates into the detection function which is used in the density estimation calculations.
1 Like
I will try to explain one of the assumptions in distance sampling, Accurate measurements. Distances have to be measured correctly, biased measurements, if uncorrected, are problematic, and in cases when detection distances and angles are recorded, rounding of small angles to zero can also cause problems with estimation. When i was still new to the department and as an untrained observer, i was always issued with a range finder to accurately measure distance and bearing. In this case training and technology was used to ensure accuracy.
Thanks Lucy, this is helpful. I was wondering about how you get around this as so may animals would move!
Does using mixed methods, i.e. dogs and video cameras on the line but not off the line, cause other issues in terms of the modelling?
I donât think thereâs still something that hasnât been explained by someone above, but the assumption of no movement I take to understand as we must record the true location of the animal we have seen during the survey to satisfy the model. If you measure the location of the animal after it has run away this would affect the detection function. The distance would be longer, which could inflate the detection function and make a species seem more detectabile than it really is. When you then use this probability of detection in the density estimate calculation, it would affect itâs reliability, as D = number of sightings / (probability of detections * area)
@LucyTallents I also wanted to ask about acoustic data. As you have no way of knowing the distance the animal is away from the sensor, I guess this canât be used to assess abundance or density?
1 Like
DETECTABILITY: in terms of our fieldwork, I will use the example of our waterbird surveys. We observe from the riverbank whilst the vehicle moves, at a sighting we stop to collect data such as species name, coordinates for example. Some of the factors which affect detectability are how well observers can identify waterbird species, how close our vehicle can get to the riverbank, habitat that is most dominant at a particular section of the river, and strength of instruments we use like binoculars. These directly affect the recordings which are captured.