Poster: Daily pollinator activity patterns vary strongly between plants
Below is a poster I presented at the Plant Sciences Symposium hosted at OSU. I mostly went so I could talk to people who know more about plants than I do and pick their brains, but the poster was well received! I’ve been pursuing an ad hoc project where I record by any type of flower I can find and it’s turning into one of those side projects that’s more interesting than your planned research. I’ve presented on the results at various venues, but this poster is the most comprehensive and the most easily digestible format.
I hope this can serve as a point of inspiration for buzzdetect users or anyone who wants to join me in delving into the underexplored world of diel phenology!
A few thoughts that didn’t make it onto the poster
1. The reasons mustard has such an incredibly tight error are (i) all of the recorders were in the same field, so they’re measuring the same trend, (ii) huge detection rates mean huge signal:noise ratio.
2. A proper quantification of uncertainty is impossible without a proper model. These ribbons are just standard errors of detection rate from all available recorders. That’s the wrong approach on many levels: not factoring out pseudoreplication, not accounting for the nonlinearity/non-normality of detection rates (which are actually beta-distributed), not accounting for non-independence of time series. Also, if you had a dozen fields that showed virtually the same curve, just shifted earlier/later by a few hours, this would show up as error near the average peak time of the curve. That’s correct in a strict sense (you really are uncertain what the detection rate will be at, say, 1:00pm), but incorrect in a more important sense: you aren’t uncertain as to the shape of the curve, only its position. This standard error approach conflates these different types of error, but I don’t see that anything can be done about this until I can derive a model that parameterizes the curve properly. This is an ongoing area of investigation…
3. What is replication? What is the equivalence between one recorder in a field for seven days, seven recorders in a field for one day, one recorder in each of seven fields for one day, one recorder placed in a different field each day for seven days…? None of this is new to statistics, I’m sure there are decent answers to all of this, but the joint factors of time series and field study make it all a little mind bending.
4. The reason for the shapes of some of these curves is obvious, or at least reasonably suggestive. Chicory flowers close early in the day, so do the cucurbits (watermelon, pumpkin). Soybean flowers don’t open until around midday. Apparently, nectar foragers provoke increased nectar production in milkweed (Broyles 2019)—maybe that’s why the duration of foraging is so long? I now suspect the beehive trend is misleading. I didn’t have much replication here (two hives, one site) and when you break it down day-by-day, it’s not that each day is bimodal as the graph suggests. Rather, there we some unimodally early days and others that were late. This demonstrates yet another difficulty in evaluating these foraging curves. And my adage that the average is the destruction of information.
Oh, also, Affinity is pretty radical!
Anywho, give it a gaze and feel free to send any questions!