Stage 2: Assuring You Have the Correct ID

Stage 2: Assuring You Have The Correct ID

This page last modified on 02 October, 2006

Topics on this page:

The "?" Scores in the Key
SubFeatures Not Specified in The Jepson Manual
Profile Pages
Random Errors in TJM Data
Separating Out The Noise
- Efficient Search Strategy
- Noisy Search Strategy
Erroneous Stage-1 ID's - Recognizing and Correcting Them

There are several potential pitfalls for the unwary in using any search key, and this Lucid based key is no exception. I've provided four tools, all of them rolled into my Profile pages, to augment the Lucid Differences pop-up and Entities Ranked-Sort tools.

In this page, I'll present an alternative search scenario for the plant that turned out to be Hazardia squarrosa var. squarrosa, as verified by detailed analysis of a specimen I submitted to the SDNHM Plant Atlas group. Though only moderately realistic, I've chosen the sequence suggested to overview the potential pitfalls -- and to present likely ways to avoid them. This presentation is (or will soon be) backed up with links to pages presenting more details on each of the pitfalls.

The "?" Scores in the Key

As a starting point, this SearchScreen (#1) presents an alternate to the scenario presented in the Tips page Screenshot (#5).

Things to note:

According to the Plant Atlas database, both the Hazardia and Isocoma subSpecies are common near the coast in San Diego County. The plants I studied before collecting the specimen grow in several areas along the trails in Torrey Pines Reserve (North Extension).
Based on hindsight, it would have been difficult to distinguish between "Plant: Duration:" of SubShrub vs. Shrub for these plants without a detailed analysis. Therefore I assumed one would have to check both for this scenario. In fact TJM says the Isocoma is a SubShrub, while the Hazardias are Shrubs. Using just one of these, and getting it wrong would lead us down the wrong path, to be recognized after additional characters had been noted.

In the Differences lists going down the Different Features listed for this search, "Disk Flowers: Corolla Color" might seem a useful choice compared with the items higher in the Differences list. Differences Screen (#1) suggests that we check "Yellow", as is seen in my photos:

Things to note:

The Lucid "(?)" notation for the two Hazardia subspecies is defined to mean that presence of this State value for those Entities is Scored as "Uncertain". See the Lucid Help files (reached via the icon) for definitions of all the available Score values.
There were several situations calling for special annotation of Scores for some States which nevertheless indicate probable presence for search purposes. Since Lucid provides no Score values that are available for definition by the author of a key, I was effectively forced to use "?" for each of the following situations:
1. The State is so common that some TJM authors have chosen not even to mention its presence, usually to conserve space in the manual. I've used the "?" code to indicate presence here, if assured by professional botanists that the State is generally true, even if an alternative State is not mentioned in TJM. That was the situation I was surprised to find when coding the key for Hazardia regarding the disk flowers that are obviously yellow in my photographs.
2. The State is one of several possibilities that cannot easily be encoded into a list of simple descriptive terms. I've used "?" here to mean you probably should refer to TJM regarding interpretation of this subFeature for this (sub)Species.
3. The TJM description or genus-level key states explicitly that the value of this subFeature tends to be variable among specimens of the (sub)Species.
4. In coding the Lucid key, I was uncertain about my interpretation of a little-used TJM term in the description or genus-level key.
The Key Menu offers a two-state option called "Retain Uncertains", which is normally in the Checked state when this key starts up. If you click that option to UnCheck it, all items marked "?" in the entire key will be regarded as Not Present when you perform a search. If I had used this option for the current scenario, neither of the Hazardia items would have appeared in the Entities Remaining list, and the Isocoma would have remained as the only matching item.
The Lucid Help document notes that Feature: State items scored "?" are given a lower Rank than those scored as "Present", for purposes of sorting lists of Features to be recommended as "Next Best" to use in a search. This partly explains why the "Disk Flowers: Color: Yellow" option appears near the low rank end of the Differences list.
There's a good reason, which I'll introduce and illustrate shortly, for avoiding the choice of "Disk Flowers: Color: Yellow" in most situations.

SubFeatures Not Specified in The Jepson Manual

The TJM descriptions for most (sub)Species don't provide any values at all for some SubFeatures. Which SubFeatures are missing varies almost randomly from one Genus to the next. Examples for all three Hazardia "Entities" include "Leaves: Order" and "Leaves: Place".

If, after choosing "Heads: Type: Discoid" in searching for the ID (that turned out eventually to apply to my specimens) of my Hazardia example you were to choose "Leaves: Order", the resulting Entities Remaining list would no longer include any of the Hazardia (sub)Species. I grant that this particular choice would not have been likely for anyone but a beginner. But the result illustrates what could happen in a case where a user chooses a SubFeature: State pair in the key for some characteristic which seems so obvious, yet different from other plants, that you feel that surely the TJM author would have given a value. Continuing to eliminate Entities Remaining until you reach a single item would leave you with a mismatched ID -- i.e. an incorrect solution to the search.!!

Lucid provides a means whereby the user of the key (i.e. user of the Lucid Player, either when initiated with your Browser, or using the separate Player program) can be protected against ever selecting one of these "Not Specified" SubFeature: State pairs. This requires that the person encoding the key will mark all States of such a pair using Lucid's "Not-Scoped" (i.e. unspecified) code. I spent a month trying this out, and discovered that this method resulted in a somewhat error-prone key that is more difficult to use than necessary for most users.

There's a much simpler way to solve this problem, though it's best applied somewhat differently in the two stages of your search:

Stage 1 (before reaching a single "matching" item in Entities Remaining).
Refer to the Differences list each time you consider using a SubFeature for your next selection, and look for entries that say "(not scored for this feature)". Obviously, you would avoid such a SubFeature if you have any reason to think that such an entry might be the solution to your search. Annoyingly, you have to close the Differences page each time before you continue with your search.
Stage 2 (after reaching the single matching (sub)Species).
Switch to the Entities >> Sorting Mode >> Ranked list on the right side of the key. Screenshot (7) at end of my Tips ... page illustrated this. Then use my Profile page for any (sub)Species you consider may possibly be a correct match for the sample plant. This page provides several bits of guidance designed to help choose the most advantageous SubFeatures for the remainder of your search. One or more Profile pages can remain on-screen for reference after being opened - they don't prevent making further selections in the main page of the key.

Profile Pages

Here's a small section of the Profile page for Hazardia squarrosa var. squarrosa that I've snipped out to provide an example for describing the kind of information provided (Use the link to pop up the whole page in a separate window. Adjust its width and height for convenience.).

Things to note:

All Features and SubFeatures in the key are listed, and appropriately indented.
The notation "Not Specified in JM" means that no information is available for any State in the SubFeature -- i.e it's neither specified as "Present" from the TJM description or genus-level key, nor marked "?" to indicate probable presence under one of the conditions listed earlier in this note.
For every SubFeature that contains one or more States marked "Present", only those States and their values are listed.
Each State that is Present has a number, as in "[22]" Spined. This number indicates how many (sub)Species "Entities" in the entire key are marked as "Present", "Rarely Present", or "Uncertain" (i.e. "?") for that State.
Some States, as in "Leaves: Margin ... [69 (1:3)] Toothed", show a pair of numbers enclosed within parentheses following the entire-key count for the State. The second number shows how many (sub)Species Entities are included in the Genus (within the range of this key). In the example, the key includes 3 instances of the Genus Hazardia. The first number shows how many of the (sub)Species Entities within the Genus group do NOT Match the State shown. In the example, Hazardia orcuttii "Leaves: Margin: Entire" is present, but not Toothed -- i.e. it does not match.

Random Errors in TJM Data

A. Radar Analogy

To set the stage for this section, I'll borrow a simplified analogy from radar. To detect the presence of some target object at a distance, a strong pulse is transmitted (using a narrow beam antenna to establish the direction), and the radar measures how much time is required for a weak echo to return to the receiver. The strength of the echo signal is determined by the size and shape of the target (and usually by the material of which the target is made), and by the distance to the target. The smaller or further the target, the weaker the echo signal.

An ever-present background of random noise is usually what limits the ability to detect the target. You probably are familiar with random noise in the form of a constant background of faint white flashing dots (a.k.a. "snow") on a Television screen when viewing a weak distant station. The noise comes from a variety of sources ranging from the electronics of the radar equipment itself to faint signals picked up from gas particles in the inter-stellar medium of our galaxy and beyond.

It's quite common for the strength of the target echo signal to be barely comparable to the strength of the noise.!! In this case, the radar display still can show presence of the target by (effectively) computing the summation of received signals following a series of transmitted pulses. The noise signal received following each transmitted pulse can be thought of as a series of 1 and 0 samples, where the index number of each sample represents a unique range (i.e. distance) from radar to the target. All signals weaker than the noise average will be represented by 0's in this series. All those stronger than the noise average will be represented by 1's. (This extreme instance of analog to digital conversion is often used as an effective approximation to more accurate sampling, when the electronics available for the conversion are not fast enough to provide more precision.)

One might compute the summation of noise signal series for, say 100 transmitted pulses, in each of many brief experiments. The total in all range indexes other than the one corresponding to the target should be roughly 50, but might vary from one such experiment to the next from about 40 (i.e. 50 minus 10) to about 60 (50 plus 10). The significance of the +/- 10 range here is that the variability of such a summation is proportional to the square-root of the number of samples added.

We might assume that the strength of the target echo is about equal to the average strength of the noise, but that the echo signal is added to the noise signal at the target's range index. That should result in a summation value of about 100 plus/minus 10 at the target range. So the target echo average is about twice the noise average, while the signal:noise ratio is 5:1 (when expressed in these crude terms).

B. Noise in the Asteraceae Key

Each of the great majority of descriptive terms used in The Jepson Manual is subject to some variability of interpretation. For example, TJM authors use at least 5 different terms commonly used to describe varying degrees of "hairy" - ness on the surfaces of a plant, and many authors use less common terms meant to convey similar meanings in this range. So there are at least the following sources of random differences (i.e. error noise) that could apply to the interpretation a user of the key might make in selecting a State value to describe a specimen plant:

TJM author variation in vocabulary used to describe any single (precisely defined) characteristic.
TJM author variation of interpretation regarding the defining plant descriptions in the research literature.
Key user variation of interpretation regarding vocabulary used in TJM.
Key user variation of judgement regarding imprecise descriptive terms used in TJM.
Key author/coder errors arising from efforts to interpret the less common TJM descriptive terms in such a way as to avoid confusing overlap among similar State terms.

Since the great majority of the Scores in the key are either "1" meaning the State is "Present", or "0" meaning the State is "Absent" (i.e. not Present), the similarity with my radar analogy becomes apparent. While different in detail, the States applying to Numeric subFeatures in the key are also subject to similar noise fluctuations.

In this key, the intensity of the noise background depends upon the SubFeature: State pairs that the TJM author (responsible for describing the (sub)Species of your plant) has chosen to include for publication.

Separating Out The Noise

In this section, I'll compare 2 strategies to illustrate use of the key for a convincing conclusion that the photos I've labelled as Hazardia squarrosa var. squarrosa are in fact what I claim. Both strategies depend on using the Lucid Entities Ranked Sort together with one or more of my Profile pages - with both displayed concurrently.

Efficient Search Strategy. Eliminates the mis-matched (sub)Species by minimizing the noise background.
Noisy Search Strategy. Demonstration of the inefficiency that results from ignoring the noise problem in your search.

Both strategies might lead you to a possible discovery that your single 100% matching Entity from Stage 1 of the search is in fact wrong. This should happen at a relatively early step in the Efficient Search Strategy -- considerably later in the Noisy Search Strategy.

The illustrations for both strategies start with the assumption that we've reached the 100% match in Stage 1 as follows:

Next step is to display the Profile page for Hazardia squarrosa var. squarrosa by clicking the little documentation icon in the top line of the Entities Remaining panel. Use this link to display that Profile in order to follow this discussion. (You'll probably want to adjust the size and shape of the window that pops up to show the Profile.)

1. Efficient Search Strategy

Begin by setting "Matching Method" to "All States" in the Key Menu.

1a. Visually scan down the Profile list looking for State items showing relatively low full-key count numbers and also matching characteristics of your specimen plant. Example: [22] Spined for "Leaves: Tip" in this example.

Repeat step 1a as many times as needed to increase the number of mismatching States chosen to 5 or more. The number of mismatching states (confusingly) is indirectly indicated by the percentage displayed for the Entity. For example if the number of Feature:States Chosen is 5, as in the lower left panel above, and the percentage is 80%, then 1 State is mismatched. If 60% then 2 States are mismatched, etc.... Additional confusion results if you check 2 or more States under the same SubFeature, because the number shown in the heading bar of the lower left panel continues to show only the number of SubFeatures with at least one checked State, i.e. without recognizing that two or more States may be checked.

For this strategy example, I checked the following additional States in the order shown:

Phyllaries: Tips: Reflexed or Recurved
Disk Flowers: Corolla Color: Red or Red Tinged -- I unchecked this after close inspection of photos showed apparent red stripes are on filaments, not petals.
Pappus: Color: Tan -- see my third photo, and first fruit photo
Pappus: # Number of Bristles/Scales/Awns: 33 -- counted from photograph
Phyllaries: Phyllaries (#) Number: 55 -- estimate based on counting one side in photograph

At this point here is the resulting Ranked Entities Remaining list:

Things to note:

The one Entity with fewer than 5 mismatched States, below the presumed 100% matching item, is a close relative -- i.e. it too is a Variant of the Hazardia squarrosa Species. In general, this pattern will emerge in situations where the key encodes data on 2 or more SubSpecies of the same Species. At this point, you could compare the Profile pages for the two subSpecies to verify the correct ID. But the more efficient approach would be to check the Genus-level key in TJM to find that the most significant difference is the expected number of disk flowers. In the plants I photographed, the number of disk flowers ranged from about 18 to 28, and averaged about 22. Setting Disk Flowers: # Disk Flwr: 22 increased the mismatch count by 1 to 6 for many (sub)Species, except Isocoma menziesii var. vernonioides, where the count remained at 5. The Genus key, and/or Profile, for Hazardia squarrosa var. grindelioides indicates 9 to 16 is the expected number of disk flowers.
The gradual drop from 40% to 30%, and eventually to 0% if one scrolls the list several panel-heights further, is a rough measure of the background noise in the data. We'll revisit this concept after comparing with the "Noisy Search Strategy".

2. Noisy Search Strategy

Restart the Key, and again match the State choices made at the end of Stage 1.

2a. This time visually scan down the Profile list looking for State items showing relatively high full-key count numbers and also matching characteristics of your specimen plant. Example: [111] Sessile for "Leaves: Attachment" in this example.

For this strategy example, I checked the following additional States in the order shown:

Phyllaries: Shape: Linear-Lanceolate
Phyllaries: Series: Several, Graduated
Disk Flowers: Corolla Color: Yellow ??
Pappus: Kind: Bristles

At this point here is the resulting Ranked Entities Remaining list:

Things to note:

As a broad approximation, the average of over 60% of the 10 chosen States still shown as matching is nearly twice the equivalent "Noise" level that remained at the same point using the Efficient Strategy.
7 of the top candidates, other than the closely related Variant of Hazardia squarrosa, still differ from the presumed 100% matching item by only 3 States. In the Efficient Strategy example, only 4 of the top candidates differed by 5 mismatched states, and all the rest differed by more.
Use of the Key-wide counts to judge which Numeric States may be advantageous for narrowing the search isn't necessarily helpful, because the Profile gives no information on the extent to which the Min:Max range for each (sub)Species overlaps the ranges of all the others coded for the same State. (I haven't yet found a concise way to represent this overlap meaningfully in the Profiles.) The Key-wide count number shown for Numeric states is a count of all Entities for which some overlap exists with the number range of the profiled Entity.

Erroneous Stage-1 ID's - Recognizing and Correcting Them

By now you should recognize that there is a risk that the initial unique 100% matched Entity you have reached in Stage 1 of your search may be wrong. If true, that immediately raises questions about the wisdom of basing your Stage 2 search on verifying that your specimen plant matches various of the States listed in the Profile page for that Entity. In fact, you should quickly begin to recognize that various Characters of your specimen don't match the States in the candidate Profile very well -- as I will now attempt to illustrate.

To set the stage, (still working with specimens of the plant that eventually turned out to be Hazardia squarrosa var. squarrosa) I started from the selected States shown in SearchScreen (#1) at the top of this page. I tried to assume the point of view of an observer who is not yet familiar in detail with the plants of the Torrey Pines State Reserve (where my specimens were growing). On looking through the Differences page, I might well have decided that the "Leaves: Surface" item, which is third on the Different Features list, would offer a choice leading to a single match, i.e. Isocoma menziesii var. vernonioides. This subspecies is very common in the nearby area (as verified by a check in the SDNHM Plant Atlas database), and from a casual view looks very similar to H.s. var. s. While the Pleurocoronis option, suggesting the State Glandular, might be more apt to describe the leaves seen in my 8th photograph, Pleurocoronis is a desert plant that does not occur in our area. The leaves seen in the 9th photo could well be interpreted to have a Scabrous (Rough to Touch) surface.

DifferencesScreen (#1b) illustrates this situation:

And SearchScreen (#2) shows the result after selecting Leaves: Surface: Scabrous.

On scanning through the Profile page for Isocoma menziesii var. vernonioides to look for appropriate subsequent clearly observable State choices, we note the following:

Under "Stems: Surface:", none of the available choices matches my photographs, i.e.:
- [7] Scabrous (Rough to Touch)
- [21] Striate, Grooved, or Ridged
- [106] Glabrous
Under "Inflorescence ... : ... Arrangement:", the one choice doesn't match:
- [34] Cluster Rounded
Under "Fruit: Surface:", the one choice doesn't match:
- [95] Hairy

These discrepancies are enough to suggest that it's time to introduce the All Characters Profile page, which you'll find under the pseudo-genus and species names "ZZZZZ" - so named to place them at the very end of the Entities tree or lists. This page lists the key-wide counts for all States used to characterize plants.

Scanning down this Profile, look for low-count States that are appropriate to the plant specimen, which when selected reduce the Rank percentage for the suspected 100% matching Entity, while at the same time raising the Rank percentage for one or more other Entities near top of the list. For example, "Phyllaries: Tips: [30] Reflexed or Recurved" is an obvious choice, giving the result in SearchScreen (#3):

Things to note:

The Entities Remaining count has dropped to "0" because no Entity is now listed at 100%.
The Isocoma m. var. v. Entity has been demoted to 1 mismatched Feature.
The two Hazardia subspecies remain at 1 mismatched Feature each.
While the Pleurocoronis species remains at 1 mismatched Feature, we already know that it is not appropriate because of location, and expect that it should not be promoted eventually to 100%.
The three other Isocoma subSpecies previously shown at 80% matching have been demoted from 1 to 2 mismatched Features each.
Except for the Cirsium and Psathyrotes Entities, all others previously showing 2 mismatched Features have been demoted to 3.

While you could continue to scan the All Characters Profile looking for additional likely States to select, it's probably best at this point to discover which of the seven States already checked accounts for the 1 mismatched Feature shown for the two Hazardia Entities that remain (since we already know that the other two Entities listed now as mismatching by 1 Feature are probably not correct). To do this, uncheck one State at a time while noting the result in the Entitities Remaining ranks.

In the case shown, the difficult-to-judge "Plant: Duration:" States SubShrub vs. Shrub should be high on your list of suspects. In fact for the Hazardia and Isocoma examples growing nearby in the Torrey Pines Reserve, one cannot see much difference re. this SubFeature by just looking at the plants!! Conclusion: UnCheck SubShrub. This results in the two Hazardia entries remaining at 1 mismatched Feature each, while all other Entities get demoted by 1.

Note: If you try replicating this exercise, be aware that I used Key >> Matching Method >> Any State mode. As a result both States under "Plant: Duration" did count as just one in the Features Chosen count, which is used in calculating the percentages matched in the Ranked Entities Remaining list. So the Hazardia entries remain at 83% at this point.

We still need to discover which of the 6 States still checked accounts for the 1 mismatched Feature for both Hazardia entries. "Leaves: Surface: Scabrous" seems most likely because it seemed a bit of a stretch when we thought it might have been describing the Isocoma m. var. v. candidate. Conclusion: Uncheck Scabrous. This results in promotion of both Hazardia entries to 100%.

At this point, it would seem reasonable to continue the search from the point where my description of the Efficient Search Stategy began.

Stage 3 - Differentiating Among (Sub)Species Within One Genus

--klb