Saturday, January 11, 2020

Size comparisons

In a previous post to this venue, various of us Colorado reviewers suggested how best to use the details box for species that are flagged by the relevant eBird filter. In this post, I wish to elaborate on that topic, suggesting better ways to use size comparisons to convince those suspicious busybodies that are eBird reviewers to accept reports.

Something that I see a lot as a reviewer is a phrase something like this:

… about the size of a warbler...

or

… about the size of a sparrow...

and similar entries. These sorts of comparisons are fine with some groups, but these two groups have fairly large size variation from smallest to largest, so the phrase does not really work as a good descriptor of size.

As example, I start with the warblers, that is, the New World warblers (family Parulidae). Now that Yellow-breasted Chat has been moved to its own family (Icteriidae; not to be confused with the New World blackbirds, Icteridae), the size variation within the family has been greatly decreased. However, the range of sizes of ABA-Area warbler species (using the length estimates provided in the Sibley Guide, 2nd ed. as a gross indicator) is 4.25-6 inches. Yes, that 1.75 inches from smallest to largest is a pretty small absolute difference. However, as a relative difference, it's huge: >41% of the smallest length.

If we take that 41.2% size difference in warblers and scale that difference to something larger, like humans, then that relative size difference becomes more apparent. Using data from this website, the average heights of human males in various countries ranges from 160 cm (5'3") in Timor to 182 cm (just less than 6'0") in The Netherlands. That difference of just about nine inches is 13.75% of the smallest average height. As you can see, the relative difference in size across warblers is MUCH larger, in fact, just about three times larger!

So, is your bird the size of a Lucy's Warbler (4.25"), the size of an Ovenbird (6"), or of something somewhere betwixt the two, say something common like a Yellow Warbler (5")?

The sparrows provide another, even more-pointed example. The range of lengths of ABA-Area sparrow species (family Passerellidae; not to be confused with the Old World sparrows, such as House Sparrow, family Passeridae) is 5-9.5"! That is a difference over the smallest species of 90%, more than double the difference among warblers! Of course, if we remove the towhees from consideration, that difference changes to ONLY 50% (5-7.5").

So, is your bird the size of a Grasshopper Sparrow (5"), a Vesper Sparrow (6.25"), a Fox Sparrow (7.5"), a Spotted Towhee (8.5"), or an Abert's Towhee (9.5")?

The take-home message is to be more precise in your size estimates where possible. That is because we reviewers don't know what you mean if you compare something's size to that of a large, variable family of birds. Imagine comparing some bird to a family like Accipitridae, which ranges in size in the ABA Area from male Sharp-shinned Hawk to female Bald Eagle.

Tuesday, December 31, 2019

"Unmistakable"

Many species of birds occur in Colorado and Wyoming that some observers might consider "unmistakable," such American Avocet, American White Pelican, and American Robin. However, as anyone that has spent any time reviewing eBird reports or browsing the incredible collection of photos submitted in eBird checklists knows, there probably is no truly unmistakable species. Below, I provide some examples.

1) Viewing conditions can make IDs difficult and many inexperienced and less-skilled observers can have difficulty distinguishing among Snow Goose, American White Pelican, and Whooping Crane -- all large white birds with black in the wing.

2) Some species might be "unmistakable" in circumscribed conditions, but not in a larger sense. In Colorado, Brown Thrasher is a widespread, though local, breeding species on the plains. As such, there are few species with which it might be confused. However, during migratory periods, any number of largish, ground-foraging passerines might be -- and have been -- mistaken for the species, particularly Hermit and Wood thrushes and Fox Sparrow (Red). The species is fairly rare in the state in winter, so care should be taken in identifying those birds. Additionally, there are four accepted Colorado records of Long-billed Thrasher, three in winter, so it behooves the observer of a reddish, streaked thrasher to consider that possibility at that season. That differentiation requires fairly good views, and details of even Brown Thrashers in winter should include details on how Long-billed was ruled out.

3) Some observers, mostly new birders, make incredible identification mistakes. While I could enumerate a panoply of mistakes that would cause amazement to experienced observers, I will provide just one that should serve to prove the point. An Aplomado Falcon report was submitted to eBird from a California observer. From the details provided, that person was a new birder. As the observer thought that it was some type of raptorial bird, the observer had asked a falconer friend for his thoughts on the bird's ID, and he had responded that the orangey-red coloration below left only Aplomado Falcon as an option. Unfortunately, the nicely photographed species was a Red-breasted Nuthatch.

The above points illustrate why eBird reviewers cringe when an eBirder uses "unmistakable" as part or parcel of the comments in which details of the identification should be provided. Please avoid using that term.

Sunday, December 1, 2019

Details provided about counting relatively large numbers of birds

The eBird help section has an excellent article on counting large flocks, passels, gobs, and masses of birds, and I encourage all eBirders and other birders to read it. This article, however, is about the details provided by eBirders to support their submitted large numbers of birds that the relevant filter flags.

In the above-linked article on the details box, is this sentence:

For entries that are flagged due to a high number, a simple comment on how the number was derived is what we want:  "counted 1x1," "counted by 10s," or "estimate" usually suffice.

I wish to add a corollary to the above. For numbers of a species that flags at a low value, say <10, particularly for numbers <5, the careful eBirder may also want to provide some ID details. That is because such low filter limits are created by one of two aspects of occurrence: 1) The report date is near either the front end or back end of a migration-abundance taper (see Fig 1 here) and higher numbers might actually involve other species (such as Clay-colored or Brewer's sparrows among Chipping Sparrows) and 2) the species is typically found only in low numbers within the relevant filter region.

The main thrust of this article, though, is points about methods of counting, as reported in species comments by eBirders.

Point 1: Many people seem unaware of the general mathematical concept of "Greatest Possible Error." I learned this in math class in either elementary school or high school, and it has stayed with me throughout my life. Relative to this essay's subject matter, this concept of Greatest Possible Error means that if counting by 10s, then one should allow for an error of 5 for each block of 10. Additionally, that Greatest Possible Error can go in either direction, that is a block of 10 might hold as few as 5 or as many as 15.

Example 1: I am counting Red-winged Blackbirds as they come into a marsh to roost for the night. I have decided that the rate of arrival is low enough that I can easily count the birds in blocks of 10. The arriving flock peters out and I have arrived at 270, counting by 10s. However, I see another few birds approaching and realize that there are only 4. Since 4 is smaller than the Greatest Possible Error of my counting by 10s, I do NOT add those four birds in, either as 4 or as another block of 10. Had there been 6 more birds, I would report a total of 280.

Corollary 1a: When counting by 10s, my final total MUST end in a zero. That is, one cannot count by 10s and report 276, for multiple reasons. First, 6 does not equal 10, nor vice-versa. Second, the possible error, particularly the Greatest Possible Error, in the previous 27 blocks of 10 so swamp those measly 6 additional birds that the 6 are irrelevant. Thus, when counting by 25s, the reported total must be a multiple of 25, such as 175, 350, 6000, etc. When counting by 100s, the result must end in at least two zeroes.

Corollary 1b: When dealing with very large numbers, say >9999, the concept of Significant Digits comes into play. Considering all of the possible errors inherent in arriving at a figure of 15,017, there is no good reason to report those last 17 birds. By reporting those 17, one is entering the realm of False Precision.

Point 2: A count is a count is a count. A count is neither conservative nor liberal. If one reports on a count, either by 1s or 5s or 10s or 37s, it is simply a count. It is understood that mistakes are made during counting and that the possible sources of counting mistakes are legion. No modifiers are needed. Additionally, unless one specifies the size of the counting block, a "count," whether "conservative" or not, is assumed to be by 1s, thus is unlikely to end in zero (1-in-10 chance; though see next sentence) and really unlikely to end in two zeroes (1-in-100 chance) . There are caveats about probability when counting, as there is actually a much larger chance than 10% of a count of 1 for a number of reasons, but once one is past those first few birds, those estimates in parentheses, above, are reasonably accurate.

Example 2: In my experience in Colorado, the most difficult flocks to count are large, dense flocks of American White Pelicans. Despite that I have learned over the years that "counts" of such flocks are invariably lower than the number of birds present, I still find it impossible to arrive at an accurate count until the birds spread out or leave in such a manner that I can get a better count. However, unless the situation changes to allow a more-accurate count, I report the number that I counted, knowing full well that it is probably low. I report such, as I can justify that number; I know it to be at least the minimum number of birds present. That probably-low count is NOT a "conservative count," as I counted each apparent separate individual that I could detect. Who knows, my count might be bang on, so, again, it is not a "conservative count," it is simply a count by 1s (or 1x1 -- "one by one" -- in Van Remsen parlance).

Corollary 2: As an eBird reviewer, I have seen many estimates of American White Pelican numbers that were low or crazily low, sometimes accompanied by photos that prove that the estimate was low. This happens frequently in late summer and fall when the species masses in large numbers at particular water bodies on Colorado's plains. In fact, reported counts of American White Pelicans at places such as Jackson Reservoir, Morgan County, can differ by an order of magnitude, that is 10x larger or 1/10 the number (and don't get me started about the illogicality of the phrase "10x smaller") -- on the same day! At approximately the same time!

The take-home message of this essay is that care should be used in counting and in reporting of numbers of birds.

Saturday, February 23, 2019

Getting Information From eBird

Figure 1. Average counts by week on the public eBird checklists of Cackling (blue line) and Canada Geese in ten Colorado counties straddling the I-25 corridor from Larimer and Weld counties south through Pueblo County. Data are from Larimer, Weld, Boulder, Broomfield, Adams, Arapahoe, Jefferson, Douglas, El Paso, and Pueblo counties. In eBird, "weeks" begin on the 1st, 8th, 15th, and 22nd, with the last week of each month incorporating however many days are in the month after the 21st. Thus, the range of time in that last week spans 7.25-9.00 days, depending on the month. Counts of Cackling Goose in the area considered span the range 0-115,000, while counts of Canada Goose span the range 0-20,000. All 366,606 public eBird checklists (as of 23 Feb 2019) from the defined area were used to generate this chart. Click on figures to see larger versions.


While many eBirders know at least the basics of entering their data into the eBird database, I have found that relatively few have much in the way of knowledge about extracting information from the database, such as that presented in Fig. 1, above. That figure shows quite graphically (pardon the pun) that Cackling Goose greatly outnumbers Canada Goose nearly the entire time that the species occurs regularly in the area, a fact that many birders seem not to know. Since these sorts of data products are the entire reason behind eBird's existence, this post illustrates the ease with which anyone may extract such. The results of eBird data mining can assist in understanding of occurrence patterns of birds in many parts of the world.

While occurrence graphs such as that presented above might seem difficult to make, the task is really quite simple, as the eBird output engine does all the heavy lifting. All one needs do is to tell eBird what one wants to see in the way of such data presentations, assuming that the facility exists in eBird. The rest of this post will present a visual step-by-step instruction of the process. Reading these instructions will take much, much longer than the actual output process takes. Note that I include a screen capture of each step in the process, with examples for most steps, of both computer (laptop or desktop) and phone (iOs version only) screens. In all of the figures, the phone example is on the right side (if present). Additionally, pink arrows are included to direct your attention to the next step in the process.

STEP 1


Figure 2. After going to the eBird home page (ebird.org), one will see the important bits as presented in the figure. On a computer, click on the Explore link; on a phone, tap on the menu icon in the upper right and then tap on Explore.

STEP 2

Figure 3. On the Explore page, click/tap on the Bar Charts link. (Say "hi" to Marshall Iliff of eBird HQ.)

STEP 3

Figure 4. Select a country (top arrow) and, if necessary, a state/province (bottom arrow). In the computer version, I have already highlighted the state, which then fills that state name in the right column that allows choosing the entire state or just select areas (you can select as many as 15 counties; this is where I selected the 10 counties that I used in Fig. 1). On a phone, you will have to tap the state/province scroll bar open and find the state/province that you want. Once you have selected the state and which of the four options in the right column, scroll down and click/tap the Continue button. Note: The phone version of this figure shows other options for areas that are not shown in the computer version, but only because they do not fit on a single screen as configured. One of these is the My Locations option. This is where you can get a summary of your yard list, should you want such. (And that is why you should always have a single location that you use for entries from your yard!)



STEP 4

Figure 5. To select counties/provinces on a computer, hold down the Ctrl key and click the individual counties that you want. On a phone, scroll through the counties list and tap each one that you want. Once your selection is complete, click/tap the Continue button.


STEP 5

Figure 6. Note: The computer and phone screens are identical for this step, so I present only one. The result of hitting the previous Continue button is this long-scrolling page showing the occurrence bar charts for every species and non-species entry recorded on public eBird checklists for the area of interest (in this case, the ten counties selected for the graph depicted in Fig. 1). You can see that the selected counties are listed. Careful observers will note that I forgot to include Denver County here and in the county selection for the graph depicted in Fig. 1! Oh, well. The arrow points at the column of graph icons that you will be using in this step. Scroll down to one of the species that you want to investigate and click on that species' graph icon. For this example, we will stay with the goose investigation, so you might want to click on the graph icon for Cackling Goose. Note: All of the actual species entries are hot links, which take you to the respective eBird species page.


STEP 6


Figures 7a and 7b. This page shows three graphs (you might have to scroll if you view the page on your computer in landscape mode) pon clicking on the Cackling Goose graph icon in the previous step. The uppermost graph is a bar graph (in green) and the second is a line graph (in blue). Both show the frequency of detection of Cackling Goose on the public eBird checklists of the region of interest. The bottom graph is a bar graph showing the weekly sample size of public checklists informing on the line graph, above (and recall that those individual weekly sample sizes total the all-time sample size of 366,606 public checklists from the ten counties selected, as noted in Fig. 1). Though your eyes might immediately go to the line graph, both frequency graphs and the sample-size graph are important. However, at this step, what is truly important is presented in Fig. 7b, the arrow pointing to the Change Species button. Click/tap that button.


STEP 7


Figures 8a, 8b, and 8c. Once you click/tap the Change Species button in the previous step, you will see the screen presented in Fig. 8a. If you were actually going to change the species, you would un-click the check box in front of Cackling Goose. However, since we want to add another species to the graph, we need to keep that check in that box and then start typing the name of the species that we want to add, here being Canada Goose. Once you have the species that you want, click/tap the Continue button. Note: A somewhat shorter method of getting to the species that you want in this box (or any other species-selection boxes in eBird), one can use the four-letter code for the species, if you know them or can figure them out.

STEP 8

Figure 9. After clicking the previous step's Continue button, we are back to the screen showing three graphs, as in Step 6. This figure presents only the line graph from that screen; note the key indicating which species is which. (In case you have problems discerning the colors, Cackling Goose is represented by the lower line.)

Now that we've gotten both species that we wanted to investigate into the graphs, it is time to explain the various data-presentation options. The pink arrow in Fig. 9 indicates that of the six data-presentation options, we are looking at the frequency data. This graph indicates that even when both species are present in the region of interest, Canada Goose is reported more often, almost certainly because it is found at more locations in the region. To get to the data presentation that is presented in Fig. 1 (at the top of this post), we will need to click on the Average Count tab (at the opposite end of tabs from the Frequency tab). If you do not understand what a particular tab presents, you can always click on the "What is..." link in the bottom-left of the screen for an explanation. Note: On the Frequency tab, the sample-size graph (the bottom graph) presents, as noted in the caption to Fig. 1, the number of public eBird checklists from the region of interest and throughout the time period of interest. Note: The sample-size checklist on all other tabs presents only the number of checklists on which the individual species was/were recorded (thus does not include zero counts).

Finally, once onto the screen presented in Fig. 9, one can change the time span used, the species, and/or the region of interest. The ease with which you can change the facets makes it easy to compare occurrence patterns of different areas -- say, eastern plains vs. West Slope, different time periods -- say, the 1990s vs. the 2000s, or, as here, a couple species of geese (one can compare as many as five species). You can also get at this facility from the Region option on the Explore tab.

Friday, June 29, 2018

Providing details on flagged high counts of birds

Home-page screen of the current iOs beta version of the eBird application (app). Click on image(s) to see larger versions.

In the beginning, the eBird app required users to provide details for any species that the relevant filter flagged as unusual, either due to the species' rarity or the count being higher than the filter limit, else the checklist would not submit. In 2017, eBird Central decided to dispense with the requirement for details on high counts in order to submit checklists, despite the moaning and gnashing of teeth by many eBird reviewers (including the entire Colorado contingent). This change was incorporated into the then-new Android version of the app and is in beta-testing mode in the iOs version. While both versions of the app still require written details in order to submit reports of rarities, reports from the Android version do not require details for non-zero counts exceeding the filter limit, and the iOs version is heading that way.

Many eBirders have probably rejoiced at this change, but we reviewers really dislike it. That is because a large percentage (though still a minority) of reports in our eBird review queues have no details due to this change. That means that reviewer workload got substantially larger and for very little benefit to eBirders in general. Before the change, something like a simple "estimate" or "counted by 1s" or "counted a portion and extrapolated total" were all that we requested for reports of high counts. The primary reason that reviewers want those details is so that we can be sure that the number reported was the number intended. Even on a phone, it is possible to make a mistake (or series of mistakes) and report unintended numbers.


Because filter limits are kept fairly tight in much of Colorado and Wyoming, reviewers in this region already have a larger task than do many, so this extra workload of having to query observers for details on high counts can be the straw that broke the camel's back. That broken back might result in blanket movement of non-documented reports of counts greater than the filter limit into eBird's non-public data, and none of us -- eBirders or reviewers -- want that. So...


PLEASE REPORT DETAILS OF HOW ANY HIGH COUNTS WERE DERIVED. (Again, see here.)


For those that haven't studied how the new app versions look and act, below are some illustrated examples of what your app can show you and what we, as reviewers, would really like eBirders to do.




Figure 1. iOs beta version app showing a hypothetical checklist. Note that the Rock Pigeon tally is indicated as exceeding the filter limit (which is set at 39). Also note that the Inca Dove entry is of a species that has a filter limit for the date and location of zero or is not on the relevant filter at all (in fact, it is not on the filter, as there are no county records). In order to submit the checklist to eBird, the app will require comments for the Inca Dove. However, it will not require comments on the high count of Rock Pigeons. We repeat:  PLEASE REPORT DETAILS OF HOW ANY HIGH COUNTS WERE DERIVED.


Figure 2. Upon tapping on the Rock Pigeon entry in the checklist on the app (see Fig. 1), one can add the detail that we reviewers would like to see on how the high count was derived. We do not need a novella written, just a simple, brief description of your counting/estimating technique (here, "estimate").

Figure 3. Unlike for the Rock Pigeon high count (Fig. 2), rarities require a bit more text. Minor rarities or those entries of species that are only marginally early or late can be supported by skimpier details, but true rarities, like this Inca Dove, need to be supported by firm details. This example is just barely enough for validation, though the promised photos would obviously help a lot.


Figure 4. We reviewers, all being eBirders ourselves, know what a pain it is to have to type much more than a few words into the comments field. However, assuming that one follows through and provides those promised details and/or photos, simply letting us know that you understand why a species flagged and that you'll get to providing those critical details as soon as you can get to a more-typing-friendly machine serves as a useful placeholder. However, PLEASE PROVIDE THOSE DETAILS AND/OR PHOTOS AT YOUR EARLIEST POSSIBLE CONVENIENCE so that we do not have to spend the time querying you.

[A side note about numbers: If one counts a flock of birds by 10s, one cannot report 362 (even if one counted 360 and then two flew over later), as that number is not divisible by 10 without figures to the right of the decimal. It's kind of difficult to have 0.2 of a live bird. One also cannot "count by 5s, then 10s" and report 365, as one's total must be a multiple of the grossest sample size -- in this case, 10. Think of it like this: Before starting a road trip, you check your car odometer, which reads 26362. Upon arriving at your destination, you check it again and it reads 26398, yet you report to the friend that you are meeting that you drove 36.2 miles. The process of counting in smaller sets of birds is similar.]

Tony Leukering and Scott Somershoe (CO reviewers)