Hi Rebecca,
Thanks for your comment. I noted those parts but they don't fully address the
issue that I described. This isn't exactly in my wheelhouse either, but I deal
with some similar issues in sampling of groundwater contaminant plumes.
The elimination of checklists was used to screen out locations with too few
ebird checklists (i.e. low temporal density of observations) but this does not
address the problem of locations where observer density has increased.
The logistical regression seems to be aimed at determining the median first
arrival date for the remaining set of observers in an area (the author calls
this "mean first arrival date" but the procedure that he describes should yield
an approximation of the median rather than the mean).
If all observers were independent, then this median value *could* be seen as
comparable, in some average sense, to the first arrival date recorded by a
single observer sitting in a lonely cabin by the woods, in the same place a
century ago.
But observers in the modern network are not independent (as illustrated by the
fact that we're all discussing Rufous Hummingbirds by e-mail right now). A
single report of an outlier on the leading edge of the arrival curve,
communicated through eBird and other digital means, triggers an intensified
focus on that species and its expected habitat.
In the case of Rufous Hummingbirds, I'd guess that everyone reading this spent
a little more time watching their feeders after they saw the first reports -- I
sure did! The northward range expansion of Anna's Hummingbirds and development
of a birdfeeding industry are also confounding factors. More observers are
likely to have feeders out and filled already in February. That increases the
chance of one observer picking up an early migrant, then alerting the local
network of observers.
The "median first detection date" for a dense network of observers is going to
be conditioned on the first detection in the network, most plausibly leading to
a forward shift in the apparent first arrival date. Correcting for this bias is
difficult but should have at least been discussed in this type of paper.
The argument concerning "the expertise of historical observers" is speculative
at best, and possibly based on a faulty premise. Modern observers are equipped
with much better optics, identification guides, audio recordings, etc. as well
as historical knowledge regarding when to expect certain migrants. Along with
use of attractants such as hummingbird feeders, playback devices are now in
widespread use. This author attempts to discount these factors by citing
biographical information about a single historical participant who reported the
first Rufous Hummingbird in one locale for more than 20 years. Such a record,
if you think about it, could only be possible in a very sparse local network of
observers.
I suspect that your final remark here, "... it does seem an interesting way to
bring two citizen-scientist data sets into conversation" is a among the main
reasons why this paper was accepted for publication. It's good to see someone
trying to work with the historic phenology data, considering the diligence that
it represents. But I'm skeptical of its use for supporting this kind of
conclusion about climate-related impacts.
Joel
On Thu, 2019-03-21 at 20:55 -0700, Rebecca Hartman wrote:
Hi Joel,
If I understand the article, he tried to address this issue with the
elimination of numerous checklists based on a four-parameter best- fitting
logistic curve model. Then he argues near the end of the article that, "Given
the expertise of historical observers and their specified task of identifying
first arrivals (Zelt et al. 2012), one could argue that historical first
arrival dates would more closely approximate true first arrival dates than
extracting mean first arrival dates from eBird checklists. If this is the
case, then the comparative method described in this paper would have the
tendency to underestimate climate related advancements in birds." (p. 541)
Granted, this stuff is way outside my wheelhouse, and I have no idea how
other studies construct statistical models, but it does seem an interesting
way to bring two citizen-scientist data sets into conversation.
Rebecca
On Thu, Mar 21, 2019 at 7:13 PM <clearwater@xxxxxxxx> wrote:
I'm mildly but not entirely surprised that this paper got published in a
journal as prestigious as the Wilson Journal.
There is no correction for, nor discussion of observer density, which is a
key factor affecting the studied variable, "mean first arrival date."
If you consider detection of first arrivals for easily-recognized species
such as Rufous Hummingbirds as a random process, then with more observers
you'd expect that "mean first arrival dates" should creep toward earlier in
the year, even if the actual migration phenomenon is unchanged.
I'm a big fan and supporter of the North American Bird Phenology project that
Jessica Zelt led.
I also helped Stanford researcher Dena MacMynowski on the 2007 study cited in
this paper, by providing her with boxes full of old journal records from
Oregon Birds etc. She and her colleagues reached a much more credible
conclusion, that "first arrival dates" are not a robust estimator of the
timing of bird migration.
The timing of spring migration *may* be creeping forward in response to
global climate change. But this type of analysis is not strong evidence, even
if it supports what most of us believe.
--
Joel Geier
Camp Adair area north of Corvallis
--
Dr. Rebecca Hartman
Associate Professor of History
History Department
Eastern Oregon University
http://eou.edu/history ;
541-962-3599