"Challenges in Detection Function Fitting and Segment Width Adjustment for DSM Modeling

66 views
Skip to first unread message

Fiama Peña Lodis

unread,
Mar 22, 2025, 2:50:05 AM3/22/25
to distance-sampling
Hello to everyone and thanks in advanced for reading me!

I’m working with a dataset to evaluate the abundance of guanacos as a function of predictor variables (anthropogenic, environmental, etc.). For one sampling station, I had very few count data, which raised two concerns: first doubt, when fitting the detection function using the entire dataset (27 data points), the fit was very poor. Following the book’s recommendations, I eliminated the 10% of the most distant data points, leaving me with 24 data points. The detection function fit decently, although it's not perfect, and this adjustment resulted in my segments having a width of 1350 meters. Second doubt, increasing the segment width makes it much larger than in other sampling stations, but at the same time, it reduces the number of segments. I know this is a problem when modeling with DSM?.I understand I need to make a decision to run the model. What would be the least problematic approach?

Eric Rexstad

unread,
Mar 22, 2025, 8:35:04 AM3/22/25
to Fiama Peña Lodis, distance-sampling
Fiama

Welcome to the list. You are being very conscientious thinking about truncation and its effect upon segment area.

Recognise that the small number of detections will have at least two consequences: a) your estimate of detection probability will be compromised because the detection function model fitting will be speculative and b) the number of segments containing detections will also be small also compromising the predictive ability of the model-based inference regarding predictors influencing guanacos.

Those two challenges overshadow the challenge associated with enlarged segment sizes. Guidance about segment size is just that, guidance. Have a look at the following, where authors examined differing segment sizes and the effect of those sizes upon their inference
  • Redfern JV, Barlow J, Ballance LT, Gerrodette T, Becker EA (2008) Absence of scale dependence in dolphin–habitat models for the eastern tropical Pacific Ocean. Mar Ecol Prog Ser 363:1-14. https://doi.org/10.3354/meps07495

One of their conclusions was
The product-moment correlation coefficients between densities predicted at the different resolutions were quite high (Table 4), with average values ranging from 0.77 (Risso’s dolphin) to 0.92 (striped dolphin).
That conclusion won't necessarily hold for your system, but it is one of the few empirical examples where the question of segment size was investigated.

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Fiama Peña Lodis <fiam...@gmail.com>
Sent: 21 March 2025 19:49
To: distance-sampling <distance...@googlegroups.com>
Subject: {Suspected Spam} [distance-sampling] "Challenges in Detection Function Fitting and Segment Width Adjustment for DSM Modeling
 
Hello to everyone and thanks in advanced for reading me!

I’m working with a dataset to evaluate the abundance of guanacos as a function of predictor variables (anthropogenic, environmental, etc.). For one sampling station, I had very few count data, which raised two concerns: first doubt, when fitting the detection function using the entire dataset (27 data points), the fit was very poor. Following the book’s recommendations, I eliminated the 10% of the most distant data points, leaving me with 24 data points. The detection function fit decently, although it's not perfect, and this adjustment resulted in my segments having a width of 1350 meters. Second doubt, increasing the segment width makes it much larger than in other sampling stations, but at the same time, it reduces the number of segments. I know this is a problem when modeling with DSM?.I understand I need to make a decision to run the model. What would be the least problematic approach?

--
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 visit https://groups.google.com/d/msgid/distance-sampling/8c5bea7c-662b-4b1b-978b-a0d0746dbf79n%40googlegroups.com.

Fiama Peña Lodis

unread,
Mar 22, 2025, 4:19:15 PM3/22/25
to distance-sampling
Hi Eric, 
Thank you very much for the accurate answer and the paper recommendation.

I really appreciate the information regarding segment width, but I’d like to dive deeper into another concern I have: the number of segments. By increasing the segment width ( in my case, necessary for the detection function to fit well), I reduce the total number of segments, which could increase the variability of predictor variables within each segment and negatively affect the model.

Would it be problematic to work with approximately 30 segments when fitting a DSM? How could I handle this situation to minimize the loss of precision in my analysis?

Thank you in advance for any guidance you can offer!

Fiama Peña Lodis

unread,
Mar 22, 2025, 4:19:25 PM3/22/25
to distance-sampling

Hi Eric,
Thank you very much for the accurate answer and the paper recommendation.

I really appreciate the information regarding segment width, but I’d like to dive deeper into another concern I have: the number of segments. By increasing the segment width, I reduce the total number of segments (in my case, necessary for the detection function to fit well), which could increase the variability of predictor variables within each segment and negatively affect the model.


Would it be problematic to work with approximately 30 segments when fitting a DSM? How could I handle this situation to minimize the loss of precision in my analysis?

Thank you in advance for any guidance you can offer!
El sábado, 22 de marzo de 2025 a las 9:35:04 UTC-3, Eric Rexstad escribió:

Eric Rexstad

unread,
Mar 23, 2025, 3:48:39 AM3/23/25
to Fiama Peña Lodis, distance-sampling
Fiama

There is not much more I can add to my previous email. You have few detections and few segments along your transects. You have noted the challenges here, including heterogeneity of the predictors within segments. Precision and defensibility of your estimates will be low in these circumstances.

Sent: 22 March 2025 19:35
To: distance-sampling <distance...@googlegroups.com>
Subject: Re: {Suspected Spam} [distance-sampling] "Challenges in Detection Function Fitting and Segment Width Adjustment for DSM Modeling
 
Reply all
Reply to author
Forward
0 new messages