Arabis sparsiflora - Notes on Identification Search using XID
Ken Bowles, 10 March 2007
These notes summarize my experience in using XID to search for species identification of 3 very similar plants from my photographs taken at three locations in the mountains of San Diego County. Some of the photographs may be found as the second "slideshow" sequence in the Brassicaceae family group shown on my website at http://www.kenbowles.net/sdwildflowers/.../BrassicaceaeKey.htm (Pops up a new window.)
To open the slideshow window, click this thumbnail photo in that matrix index page:
This is mainly a talking paper for use at several levels:
Detailed content of these notes will change from time to time as I learn more from communicating with various people whose roles cover a wide range in relation to these points.
Initial motivation for writing these notes arose from my attempt to explain the challenge of software design for plant species identification to a former faculty colleague who now specializes in the general problem of software for searches in complex databases. I had contacted him after recently learning about XID, and my discovery that XID offers very considerable advantages compared with Lucid and other software efforts that I've known about previously. After looking briefly at my crude photo-oriented efforts based on matrix index pages of thumbnail photos, he asked me (in effect) to provide a tangible example of how XID might help to identify one of my 4-petal example species.
I chose the example initially because it's a fairly common plant from our mountains, and one for which I have a typical, but reasonably good, series of photographs taken 2-3 years ago. Further study, while preparing this note, revealed that this species exposes quite a few subtleties that help to illustrate realities of plant identification practice as of early 2007. I'll take time in drafting this note only to hint at most of those subtleties. One broad hint: I marked my slideshow photos on my website to indicate I was somewhat dubious of the accuracy of my ID, and I need verification from a botany expert that I've gotten it right now that I've used XID to chase it down.
The sequence of XID search steps illustrated in the following are from use of the XID database covering "All known native and naturalized vascular plants" for the state of Oregon, and available on CD-Rom from Flora ID Northwest, 731 NW 5th, Pendleton, OR 97801, (Email: flora@ucinet.com). I have used only the "Flowering Plants" section of this database.Please Note: The content of all screen images in this note, if obtained using XID, is Copyright© 2007 Flora ID Northwest, All Rights Reserved.
First Steps
When it starts, the Oregon key shows a list of 3098 species in the lower left panel, and an expandable menu tree of possible descriptive "attribute" terms that are available for possible user selection in the upper left panel. The XID User Guide recommends starting the selection process with a few very obvious and visible attribute values that are likely to apply to a moderate subset of all the species in the key. To arrive at the following screenshot, I had:
As a result, the notation at top of the lower-left panel showed that 237 of the species remained available for selection, and were shown in green color in the scrolling list. Those so-far eliminated are listed below those remaining, and in red color (not yet visible while the remaining group is still large).
I then clicked the menu-heading for Flowers: Sepals: Sepal surface, which caused the array of illustrations to appear in the right panel explaining the meanings of the several attribute terms in the menu. Motivation for doing this came from several of my photos, as explained below.
To highlight the "Stellate or branching hair" menu item, I clicked over the diagram marked "Stellate hair". The number 10 in the box at left means that only 10 of the remaining 237 species have been keyed as having this attribute value. I marked this attribute value as present by double-clicking the menu item (or clicking the little "Y" icon in the menu bar). A close look at the 4th photo in my slideshow sequence shows why I chose this value. XID features two command tools for suggesting which attribute value should be selected to reduce the list of matching species significantly, and I'll be illustrating those later in this note. For this choice, I had noticed that the surface of the sepals looked somewhat "fuzzier" than one finds on most plants. That suggested that I might shorten the search because the characteristic I'd noted is somewhat unusual.
When one gets down to a fairly small fraction of all the species in the key, it's time to do a reality check after each new attribute value is selected. In this case, 9 of the 10 species now remaining are members of the Brassicaceae family - and I was aware that the vast majority of flowers in that family have 4 petals (matching the plants in my photos). If good photographs are available, then a quick glance at the key's "Gallery" of images for the species still remaining as available can be helpful. In this example, the Gallery photos for the 6 species in the Arabis genus are all quite similar to my sample of plants.
Analyze Command
Having already gotten down to just 10 remaining species, it's possible that either the "Analyze" command (Database Menu, or the Icon-bar "Eye" button) or the "Distinctive attributes" (Context pop-up Menu that appears when you right-click a species name in the lower left panel list) would give best results. Having previously more/less identified my plants (with low confidence, hence the "???" notation after the species name over the photos in the slideshow window) with Arabis sparsiflora, I looked first at the Distinctive attributes list. The options suggested there did not look promising (i.e. not very distinctive within the remaining list of Arabis species), so I went to the Analyze list as shown here:
| List of useful menus. |
XID orders this list in a way intended to place (near the top) the sub-menus most likely to shorten the remaining species list by the largest percentage. But the number of possibilities is large, and the program has no knowledge of your plant beyond the information you have already given with the attributes chosen so far. Lengths and widths are always approximate, and the known values that apply to a species cover a range of possibilities. But a quick review of the available Gallery photographs shows considerable differences of leaf shapes among the plants remaining. While the leaf shapes of most species tend to cover a range of possibilities, even on a single plant, it often proves helpful to select one from the list of likely shapes as shown here in the right panel (ignore the irrelevant "cylindrical" mouse fly-over box):
Distinctive Attributes:
When I selected Oblanceolate, the 5 species that then remained were all members of Brassicaceae. Another check of the Distinctive attributes command produced the following list (only the top portion shown to limit your need to scroll):
| Distinctive Attributes of Arabis sparsiflora/Brassicaceae/rockcress, sicklepod |
Each item marked "<1>" at the left shows an attribute that's only coded for the highlighted species in the lower-left pane. Each item marked "<2>" shows an attribute shared by the highlighted species AND one other among the group of species remaining. The list continues showing attributes for which 3 and 4 species share, since selecting any of these would have the effect of reducing the remaining list by one or more species.
I noticed that the menu "General traits of mustard family fruit:..." appears several times in this table. Characteristics of the "Fruit" (i.e. seed pods) are among the most important descriptors that distinguish species in the large Brassicaceae (i.e. Mustard) family. A quick glance at the 3rd and 5th photos in my slideshow sequence for what I had labelled "Arabis sparsiflora" shows the seed pods extending upwards (i.e. "ascending"), pointing outward horizontally ("spreading"), and pointing downwards ("drooping"). [Footnote to SDNHM folks. I now notice the genus needs to be relabelled Boechera on my website.] The next screen shot shows the result of selecting all three in the "General traits ..." menu, meaning all three attributes apply to the one species that results.
By displaying the Description page for the one remaining species, XID is notifying us that the selected attributes are a unique combination that applies to just one out of the 3098 species in the key. (The order, i.e. permutation, in which the attributes were selected is not significant.) To recap progress so far, a pop-up notice informs us that just 7 attributes have accomplished this feat. On request, we can list those attributes together as:
| Marks |
The order in which this right-panel list is displayed corresponds to the order of the menus in the upper-left panel.
Forgiveness Levels
Unfortunately, there are several ways in which this apparently unique combination may have led us to the wrong solution. For example: Along the path to this list of attributes, there have been several opportunities in which either the Analyze list, or the Distinctive Attributes list, could easily have led us to the "Stem leaf/leaflet base shape:" menu seen in this screen:
What this tells us is that the leaves are keyed as attaching to the stem either with "flaring lobes" like those of an arrowhead ("sagittate" in botanist jargon), or "clasping" (i.e. wrapped at least partially around the stem). But hold-on ... a close look at the leaves in my photographs of these plants shows the leaves attached in more like the "Squared" fashion seen in this table of diagrams. I've checked the specified attributes for Arabis sparsiflora in The Jepson Manual regarding this characteristic, and find that they match the attributes in this XID key correctly. Morover, the one photograph currently shown in the Oregon XID key shows a leaf example more similar to mine than to either "flaring" or "clasping". But several photos posted on the California statewide CalPhotos database show leaves with obvious sagittate attachment to the stem.
In fact, this apparent error is a reflection of reality and/or state-of-the-art in plant identification. The characteristics of plants of a single species do vary somewhat from location to location as a result of environmental factors, from naturally occurring hybridization or evolutionary drift, or a result of insufficient statistical sampling data included in the research conducted when the species characteristics were formally published. The practical effect of this is that there is a small, but not negligible probability that an incautious user will mis-identify the species of a plant if depending only on the first combination of attributes that leads to a single species as "target" solution.
To provide protection against this risk of errors, XID permits a user to set the "Forgiveness level" with a command in the Options Menu. In fact, after setting the three concurrent "Fruit ascending, spreading, and drooping" attributes above, I had already set the Forgiveness level to -1 in the screen seen above. This means that the combination of attributes seen in the list above will uniquely identify Arabis sparsiflora, even if any one of those selected attributes is in error (i.e. the corresponding menu item may either have a different value, or many have no value keyed at all).
Continuing the process by setting the Forgiveness level progressively to -2, then -3 in stages, we go through the same winnowing of attribute values at each stage until a single matching species is again shown. Here is one list (still accurately describing my plants) at Forgiveness level -3 that continues to lead to Arabis sparsiflora:
| Marks |
Of the 10 attribute values shown, any 3 can be wrong in this list but the resulting match will still lead to Arabis sparsiflora. To put this in perspective, the total number of descriptive attributes marked in the key for Arabis sparsiflora is approximately 119 (counting each number valued group of attributes as just one!!). Because of the probability of random variations of the keyed values, or of their interpretations by users, some moderate percentage of possible combinations of 10 to 15 attributes would incorrectly lead to single species as target solutions if the Forgiveness level feature were not invoked. I have proven this statistically with my Lucid key covering 306 Asteraceae species in San Diego County, and plan to repeat the study using a much larger XID key approaching the size of the Oregon key discussed in this note.
As a final reality check, I looked up the Plant Atlas project database locations for the Arabis genus. I used their website interface with the Berkeley Mapper application of the Google Map service. This allowed limiting the region under consideration to an area encompassing the 3 sites where I took photographs of what all look like the same plant species, as seen in this screenshot:
On this page, note that all but one of the San Diego County species in the Arabis genus have recently been moved/renamed to the Boechera genus.
As a final note, the total number of possible descriptive attribute values (i.e. other than quasi-index menu items such as Family or Genus, or Geographic location attributes, etc...) is at least 10 times as large as the number of attribute values actually keyed. That means that the vast majority of possible values has been keyed effectively as "not present", or more accurately as "no information available". The list that follows is the full list of keyed XID "data" for Arabis sparsiflora from the Oregon key:
| Arabis sparsiflora/Brassicaceae/rockcress, sicklepod |