Assessing Accuracy in Plant Identification in a
Riparian Monitoring Protocol
Fish
& Aquatic
Ecology Unit
Catherine Mae Culumber and Alison Kelly
U.S.D.A. Forest Service, PIBO Effectiveness Monitoring Project
Rocky Mountain Research Station, Logan, UT 84321
Data Subset
•Technicians misidentified 9% of specimens (26/274, excludes unknowns)
•Average Relative Importance of 16.2
•Unidentifiable specimens comprised 15% of specimens (51/351, includes
unknowns)
•Average Relative Importance of 4.2
Introduction
Two sources of error in vegetation studies are incomplete and incorrect
identification of vascular plant taxa
Unknown plants and incomplete identifications (to the genus level) can limit the
ability of researchers to draw conclusions from ecological data
•Effect of Misidentification on Wetland Rating
•While the differences were not significant, misidentifications tended to
lower the Wetland Rating slightly (Figure 6)
•of 21, 7 unchanged, 4 higher, 10 lower
•Coefficient of variation between Wetland Ratings with correct vs. incorrect
identifications was low (n=21, CV = 2.94)
Our objectives in this research were:
•describe the error associated with incomplete and incorrect identification of
vascular plant taxa
•investigate the effects of this error on data summary and analysis
Methods
100
WA
Study Site
•Interior Columbia
River Basin
•Wide range of communities
•Salix/Shrub dominated (Fig. 1)
•Graminoid dominated
•Forested (Fig. 2)
MT
90
80
OR
ID
WY
70
Fig. 3 Vegetation technician collecting data on the greenline
UT
GL Wetland Rating
NV
Results
Large Dataset
•A decrease over time in the number of unknowns collected at a sampling
reach, technicians distinguished by color (Figure 4)
•Note that most technicians consistently collected fewer or more unknowns
relative to other technicians
•General decrease over season probably due to increasing familiarity with
flora
60
50
Correct IDs
Incorrect IDs
40
30
20
10
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Site
Fig. 6. Wetland Ratings with correct and incorrect identifications for 21 sites
40
Discussion and Conclusions
35
•Number of unknowns decreased over the field season
•Possible contributing factors: an increase of flowering vegetation and
technician experience
•Indicates the importance of training and supervision by experienced
botanists early in the field season
•Consistently high or low unknowns relative to others may correlate to the
level of technician experience
Fig. 1. Salix/Shrub dominated community
# of UNKNOWNS
30
25
20
15
•Average number of unidentifiable specimens decreased over the field season,
and the data subset showed that the relative importance of those specimens
was low
10
5
0
03
20
1/
/1
10
3
00
/2
21
9/
03
20
1/
9/
3
00
/2
12
8/
3
00
/2
23
7/
03
20
3/
7/
3
00
/2
13
6/
3
00
/2
24
5/
Date
Fig. 4. Number of unknowns per day over time for 12 techs
Large Dataset
•12 similarly trained botanical technicians performed riparian vegetation surveys
using Daubenmire quadrats (20 x 50 cm) (Coles-Ritchie et al., 2003, modified
from Winward 2000)
•Technicians identified specimens to the best of their ability in the field and
collected unknown taxa in field plant presses
•Specimens were identified by experienced botanists in the lab
•Number of unknowns over time calculated
•Average number of unidentifiable specimens over time calculated
Data Subset
•To assess the rate of misidentification
•Technicians collected specimens of all (known and unknown) vascular
plants over 5% cover in randomly selected quadrats throughout field season
(Figure 3)
•Identifications compared with identifications in the lab
•The impact of misidentification on a Wetland Rating was assessed, assuming
that taxa misidentified at the sample quadrat were misidentified throughout
•Wetland Rating is an index that provides information about the moisture
availability on the streambank (Coles-Ritchie 2004)
35
30
Literature Cited
25
Brandon, A. 2003. Can Volunteers Provide High Quality Forest Monitoring Data?, Illinois
EcoWatch Network, http://dnr.state.il.us/orep/ecowatch/Data/fw/Volunteerwriteup.htm,
accessed 1/20/04.
20
Coles-Ritchie, M., R. Henderson, M. Culumber, and A. Kelly. 2003. Effectiveness Monitoring
for Streams and Riparian Areas within the Upper Columbia River Basin: Sampling Protocol,
Vegetation Parameters. Internal document. U.S.D.A. Forest Service, Fish and Aquatic
Ecology Unit, Logan, UT.
15
10
5
R2 = 0.3959
0
Fig. 5. Average number of unidentifiable specimens by day
3
00
/2
16
9/
Date
3
00
/2
27
8/
03
20
7/
8/
3
00
/2
18
7/
3
00
/2
28
6/
FHM posters home page
40
03
20
8/
6/
n
Site Wetland Rating = ∑ ( rimpi * wiri)
i =1
Where:
i = all species at a site, rimpi (relative importance) = importance / ∑ importance all species, impi = average
coveri * relative frequency, and wiri (Wetland Indicator Ranking) = value for each Wetland Indicator Status
on 0-100 scale
•A decrease over time in the average number of unknowns that were
unidentifiable in the lab for all technicians (Figure 5)
•Specimens tended to have reproductive structures later in season
•Most problematic families when sterile: Cyperaceae (sedges), Poaceae
(grasses), Apiaceae (Carrot family), Asteraceae (Sunflower family)
Avg Number Unidentifiable Specimen
Fig 2. Forested community
•Misidentification Rate
•Lower level of misidentification compared with 25% field misidentification
rate for 2002 preliminary investigation (Culumber 2002)
•Other research shows a 45% misidentification rate among volunteer
botanists in Ill., where oak and elm species were most commonly
misidentified (Brandon 2003)
•Effect on Wetland Rating
•While relative importance of misidentified taxa was high (16.2), wetland
ratings for streamside vegetation had a low CV of 2.94
•Larger dataset CV of 5.5 with misidentification as a one of several
contributing factors
•What sources of error can we control?
•Unknown specimens
•minimize with supervision
•collected unknown specimens add to data quality
•Unidentifiable specimens
•limited control with survey timing
•Incorrect identifications
•minimize with supervision and training
Coles-Ritchie, M. 2004. A Wetland Rating System for Evaluating Riparian Vegetation.
Disssertation. Utah State University, Logan, UT.
Culumber, M. 2002. Investigation of Plant Specimen Field Indentification Accuracy. Internal
document. U.S.D.A. Forest Service, Fish and Aquatic Ecology Unit, Logan, UT.
Winward, A. 2000. Monitoring the Vegetation Resources in Riparian Areas. U.S.D.A. Forest
Service, Rocky Mountain Research Station. RMRS-GTR-47.
FHM 2004 posters