distance to pan

[kaitlyngaynor]

Moving this to a GitHub issue...

From Katie:

For water, we'd discussed on the doc switching to a general "distance to water" variable, which would be the minimum distance to either the lake, a river or a pan. Looking at the available data, I think that would be determining the minimum value among the "pans_dist", "river_dist" and "urema_dist" columns - does that sound right? I know you also mentioned large pans on the document, but that variable is "pans_large_250m: Percent of area in 250 meter radius that is covered by large pans (pan layer created manually by K Gaynor and J Daskin)." As such, I don't see as easy a way to combine the different water variables if we use the percent value for pans.

The pan layer includes a lot of TINY watering holes that definitely don’t have water any time near the study period, hmm. I’m hesitant to use it… you’re right that there’s no easy way to combine the density layer and distance layers, though.

I did find the raw shape file for “conservative/large pans,” which I defined as a pan >1km2 in size. These retain water for a lot more of the year. I found a script (from 2018) that calculated distance from each camera to largest pan, but it's using R packages that no longer work and input files that I no longer have... it would probably be pretty fast in QGIS/ArcGIS, but if this isn't in your wheelhouse @klg-2016 I can figure this out in R using packages that actually work! Let me know.

Here's the shape file: Pan_largerthan1km2.zip

[klg-2016]

I have only done a single tutorial in QGIS to make a map for my mangrove work, so it's not something I think I'd be able to figure out in a reasonable time frame using that. If it's okay and you have time, could you see what you can figure out in R?

[kaitlyngaynor]

Sure thing. I just did this in this script here, using the st_distance() function in the sf package. It was pretty easy—take a look!

Here's the output (units are meters): camera-pan-dist.csv

[klg-2016]

Awesome, thank you! I'll check it out.

Is your thinking now to use this and still calculate closest "water" using distance to river, pans and lake?

[kaitlyngaynor]

Yeah I think so?

[klg-2016]

Sounds good, I'll work with that!

[klg-2016]

Update on single species covariate exploration:

• genet: we said water, termite, tree cover. none of them individually or in pairs improve upon the null model, but termite mound has the lowest AIC
• civet: we said water, termite, tree cover. all three improve upon null model (but barely for termite). water does the best among all three individually or in pairs (distance to lake does better than distance to water)
• honey badger: we said termite mound density. that model yields a singular hessian and does not improve upon the null. distance to water is by far the best fit and doesn't throw any warnings. I think it's reasonable to think that water matters for all four of these species, even if we didn't expect it especially for the honey badgers.
• marsh mongoose: we said distance to water, which does not improve upon the null (only termite mound does)

to try:

• genet: termite
• civet: water
• honey badger: water
• marsh mongoose: water

When I run this as a Type 1 (no interactions model), it beats null by 1.7 AIC points. I feel like I've read (or you've told me) that delta 2 is enough to consider models sufficiently different. What do you think?

[kaitlyngaynor]

closing distance to pan issue