Inlay

Profile

Associate prof in plant ecology at Uppsala University, Sweden. Tropical forest dynamics, trait-based ecology, natural and anthropogenic disturbance...

Bob Muscarella

Final note: Machine learning / AI approaches are being developed that may eventually help remove some of the barriers to this work. Regardless, good sample preparation and imaging is truly an art and worth the time spent! #Q-NET

Bob Muscarella

But having higher power for addressing many questions (e.g., intraspecific variation) would require more intensive sampling. This should be explicitly considered when designing a sampling plan for wood anatomy research.

Since traits can spatially vary across a sample, it's not clear how many vessels (or how much area) should be measured to get a good estimate for a given trait value. What to do?!?

We also found that the proportional area of the sample measured better explained error rates than absolute area. So, when doing this kind of work, you might consider measuring larger areas for larger samples to get a good estimate rather than measuring a standard area across sections... 🤔

New paper on sampling for vessel traits in wood anatomical studies! A short 🧵... Wood anatomical traits can provide great info on ecophysiology, ecology, and evolution but it takes a lot of painstaking work to quantify them! academic.oup.com/aob/advance-... @juliavtavares.bsky.social

Well, we had already measured ALL the vessels in a set of 164 samples across 19 species in Puerto Rico. So we thought "Let's use a digital subsampling approach to see how accuracy and precision of trait estimates varied with the proportional (and absolute area) sampled!"

So, we quantified the improvement of accuracy and precision of trait estimates relative to the "true value" (i.e., when ALL vessels were measured) with increasing proportional area sampled.

For example, quantifying vessel traits (e.g., vessel size, density), people typically collect thin cross sections and then work under a microscope to measure the features. But an individual cross section can have LOTS of vessels so people often measure a small fraction of these.

Last, we used variance partitioning to attribute variation in of various traits to across species, within species, and residual variation, with different proportional areas sampled. For broad comparisons across species, even fairly low sampling (~10% of section area) captured the main patterns.

Bob Muscarella

[Side note: In this process, we developed an R package (qwar) to process wood anatomical annotations in R. It requires an .SVG file of annotations as input (we used #QuPath to do the annotations). The package is available albeit not very well documented yet... github.com/bobmuscarell... ]

Bob Muscarella