Estimating species distributions with occupancy models

[philpatton]

Estimating species distributions with occupancy models

This would add a new example notebook for species distribution modeling with occupancy models, an extremely common task/model in ecology. See here for the relevant issue. I apologize in advance for any rookie mistakes!

• Notebook follows style guide https://docs.pymc.io/en/latest/contributing/jupyter_style.html
• PR description contains a link to the relevant issue:
  • a tracker one for existing notebooks (tracker issues have the "tracker id" label)
  • or a proposal one for new notebooks
• Check the notebook is not excluded from any pre-commit check: https://github.com/pymc-devs/pymc-examples/blob/main/.pre-commit-config.yaml

Helpful links

• https://github.com/pymc-devs/pymc-examples/blob/main/CONTRIBUTING.md

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

fonnesbeck commented on 2026-01-28T22:08:27Z
----------------------------------------------------------------

Surprised to see a linear date effect -- would have thought it to be non-linear. How is date encoded?

philpatton commented on 2026-01-29T00:44:57Z
----------------------------------------------------------------

I added a little note that the surveys took place from January to April, and also added a note saying "see below for figure showing the predictors"

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

fonnesbeck commented on 2026-01-28T22:08:28Z
----------------------------------------------------------------

Would be helpful to have a short visualization to give readers a sense of what the input data look like (related to my comment about dates above)

philpatton commented on 2026-01-29T00:45:58Z
----------------------------------------------------------------

Definitely! I added a little figure showing histograms for elevation and forest cover at quadrats, and a bar chart for the number of surveys per month.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

fonnesbeck commented on 2026-01-28T22:08:29Z
----------------------------------------------------------------

Could you not use a ZeroInflatedBinomial here?

philpatton commented on 2026-01-29T00:57:16Z
----------------------------------------------------------------

In a similar notebook on my personal webpage, I show how to use ZeroInflatedBinomial for models where $p$ is constant. I'm not positive but I don't think you can use it when $p$ varies by survey, because the zero-inflation status for sites with a detection is known. For example, if y[0, :] = [1, 0, 1], we know that the second zero is a sampling zero, and shouldn't be inflated, if that makes sense. But when detection is constant across surveys, we can sum the y's by row, such that y[0] = 2 and all zeros could be sampling or structural.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

fonnesbeck commented on 2026-01-28T22:08:29Z
----------------------------------------------------------------

Ah, okay. The linear date effect makes more sense if the surveys are only over 4 months. Should add this detail to the intro (or again, do some data viz).

philpatton commented on 2026-01-29T00:58:34Z
----------------------------------------------------------------

See comments above, but I added both the detail and the visualization.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

fonnesbeck commented on 2026-01-28T22:08:30Z
----------------------------------------------------------------

typo: Not -> Note

philpatton commented on 2026-01-29T00:58:43Z
----------------------------------------------------------------

Corrected

[fonnesbeck]

This is very cool, @philpatton. I used to build a lot of these models, so getting one into examples is exciting. Only a couple of questions.

[philpatton]

I added a little note that the surveys took place from January to April, and also added a note saying "see below for figure showing the predictors"

---

View entire conversation on ReviewNB

[philpatton]

Definitely! I added a little figure showing histograms for elevation and forest cover at quadrats, and a bar chart for the number of surveys per month.

---

View entire conversation on ReviewNB

[philpatton]

In a similar notebook on my personal webpage, I show how to use ZeroInflatedBinomial for models where $p$ is constant. But I don't think you can use it when $p$ varies by survey, because the zero-inflation status for sites with a detection is known. For example, if y[0, :] = [1, 0, 1], we know that the second zero is a sampling zero, and shouldn't be inflated, if that makes sense. But when detection is constant across surveys, we can sum the y's by row, such that y[0] = 2 and all zeros could be sampling or structural.

---

View entire conversation on ReviewNB

[philpatton]

See comments above, but I added both the detail and the visualization.

---

View entire conversation on ReviewNB

[philpatton]

Corrected

---

View entire conversation on ReviewNB

[philpatton]

Thanks for the comments @fonnesbeck! I actually knew that, because I think we're academic cousins of some sort, since my master's adviser, Krishna Pacifici, also did his PhD with Mike Conroy.

[fonnesbeck]

LGTM!

[philpatton]

Awesome! Thanks for taking the time to review @fonnesbeck!