Grantee Research Project Results

Final Report: Stream Plethodontid Assemblage Response (SPAR) Index: Development, Application, and Verification in the MAHA

EPA Grant Number: R827640
Title: Stream Plethodontid Assemblage Response (SPAR) Index: Development, Application, and Verification in the MAHA
Investigators: Brooks, Robert P. , Rocco, Brian L. , Hite, Jeremy T.
Institution: Pennsylvania State University
EPA Project Officer: Packard, Benjamin H
Project Period: July 1, 1999 through June 30, 2002
Project Amount: $397,304
RFA: Ecological Indicators (1999) RFA Text |  Recipients Lists
Research Category: Aquatic Ecosystems , Ecological Indicators/Assessment/Restoration

Objective:

The objectives of this research project were to: (1) describe the range and variability of stream plethodontid assemblage responses (SPAR) across commonly encountered gradients of anthropogenic degradation (stream acidification, forest and riparian corridor fragmentation and degradation, pollution, etc.) in the Mid-Atlantic Highlands Area (MAHA); (2) develop a SPAR-based index for use in MAHA headwaters; and (3) evaluate the reliability and resolution of SPAR by application and testing.

Summary/Accomplishments (Outputs/Outcomes):

Part I–SPAR Index: Development, Application, and Verification in the MAHA

Small headwater streams are vital components of rivers. They comprise 60-75 percent of the total stream length and watershed area in the Mid-Atlantic Highlands. The Appalachian region supports a diverse assemblage of Plethodontid (lungless) salamanders. Amphibians, in general, are considered to be valuable response indicators. Several efforts are underway to develop and test indices of biotic integrity based partly or wholly on stream salamanders.

In 2000-2002, we sampled 138 EMAP stream sites (3 km²) covering a wide range of ecological and human disturbance gradients (Figure 1). The EMAP Mid-Atlantic Highlands wadeable stream sites were originally selected by a randomized, probability-based design to allow inference on environmental condition for 184,600 km streams in the region. Stream sites cover a broad range of ecological conditions and they are situated in six U.S. Environmental Protection Agency Level III ecoregions. A large body of ecological information of these sites exists. Selected stream sites were sampled once and only at stream locations approximating EMAP stream site coordinates. Sampling along approximately 100 m of the stream channel entailed the measurement of climatic and water chemistry variables, stream channel physical characterization, and sampling for stream salamanders. Salamanders of all lifestages were captured from terrestrial and aquatic portions of three 4 m² rectangular plots (2 m x 2 m). Plots were always positioned to include dry and wetted portions of the stream channel.

 

Figure 1. Map of the MAHA Showing the Approximate Location of the 138 Environmental Monitoring and Assessment Program (EMAP) Wadeable Stream Sites Sampled in 2000-2002. Fine contours delineate Level III ecoregion boundaries (Woods, et al., 1996)

In developing indices of biotic integrity (IBIs), we used the criteria described by Waite, et al. (2000) to classify a priori each of the 138 stream sites into reference, nonreference (minimally degraded), and degraded (severely degraded). These criteria are based on the measurement of nine variables related to water chemistry, stream physical habitat, and a total macroinvertebrate count. Several other IBIs have been developed using this criteria, including fish and macroinvertebrates.

We initially screened 33 metrics. Based on this initial screening, we identified 11 metrics of potential value (see Table 1). Few of these metrics were linearly correlated to measures of degradation, but showed considerable association with benthic macroinvertebrate communities, an indication of an indirect response to degradation. Geographic and stream physical habitat was found to affect several metrics.

Natural variability was partitioned by a three-step process that consisted of the initial classification of 34 reference and near reference sites by salamander assemblage type, subsequent development of discriminant functions from environmental measurements to allow classification of new sites, and classification of new sites (nonreference and degraded) into classes identified in the first step by application of the discriminant functions. Initial classification of reference and near reference sites was performed

 

 

Table 1. Description of the 11 Stream Salamander Metrics Tested Beyond the Initial Screening and Used for IBI Construction

	Metric Name	Metric Description
A	Species Richness	Number of Plethodontid species, including woodland species (Plethodon spp)
B	No. Mountain dusky	Number of salamanders of the mountain dusky (Desmognathus ochrophaeus)
C	No. two-lined	Number of salamanders of the two-lined (E. bislineata, E. cirrigera, and possibly larval E. longicauda)
D	No. northern spring	Number of salamaders of the northern spring. (Gyrinophilus porphyriticus)
E	No. salamanders	Number of salamanders of all species
F	No intoleraants	"no. of salamanders" minues "No. two-lined"
G	No. nutrient tolerant	"No. two lined " plus number of salamanders of the northern dusky (Desmognathus fuscus)
H	No. acid tolerant	"No. mountain dusky" plus "No. northern spring" plus any woodland species (Plethodon spp)
I	No. terrestrial	Number of salamanders without gills or gill stupb
J	No. larve	Number of salamanders with gills or gill stubs
K	No. LS class 3	Number of salamanders identified as class 3 primary headwater indicator species (Ohio EPA)

by ordination (detrended correspondence analysis [DCA]) and reciprocal averaging (two-way indicator species analysis [TWINSPAN]). Discriminant functions were only developed from variables related to natural gradients. Multiple discriminant analysis (MDA) was used to develop the discriminant functions.

The classification of reference and near reference sites resulted in three classes or groupings, each consisting of 9-16 sites. The mountain dusky and northern spring salamander were indicators for Group 1, the Appalachian seal was the indicator for Group 2, and the northern dusky and two-lined were indicators for Group 3 (see Figure 2). DCA and TWINSPAN classifications were largely in agreement. Preliminary univariate analysis of the nine environmental variables revealed the three groupings stream to be related to geographic and stream physical habitat, variables previously shown to be associated with salamander metrics.

Subsequent MDA produced two mountain dusky significant discriminant functions, which were linear combinations of four variables: latitude, stream temperature, cobble cover, and stream gradient. The first and second discriminant functions accounted for 86 percent and 14 percent of the total variance, respectively. The first discriminant function was most highly correlated to latitude (r = 0.793). The second canonical function was most highly correlated to water temperature (r = - 0.659), boulder cover (r = 0.539), and slope (0.413). The second canonical discriminant function describes an environment where stream gradient becomes steeper, the abundance of boulders increases, and stream temperature decreases, as values along this axis increase. In this respect, stream sites in Group 3 are relatively warmer (or less cold), less steep, and have fewer boulders relative to sites in Group 1 and Group 2 (see Figure 3). The latter groups are similar in terms of stream habitat.

 

Figure 2. DCA Ordination Biplot Showing the 34 Reference/Near Reference Sites (Circles) and the 6 Stream Salamander Species (Crosses). Dashed lines separate clusters identified as DCA Groups 1-3. Arrows point to TWINSPAN grouping of the seven "discrepant" sites. Key: montc = Appalachian seal, Pseudt = northern red, fuscus = northern dusky, Eryc = two-lined, Gyrinp = northern spring, ochrp = mountain dusky.

 

Figure 3. Plot of Canonical Discriminant Function Scores for Group 1-Group 3. The small squares in each cluster represent the group centroids. Scores along the first canonical axis are positively and most highly correlated to latitude. The second axis is negatively correlated to stream temperature, but positively correlated to boulder cover and slope.

The overall number of sites correctly classified by cross validation by this model, with equal prior probabilities for all groups (three groups = 33 percent), was 88.2 percent. The proportion of correctly classified sites by cross validation by group was 89 percent, 94 percent, and 78 percent for Groups 1-3, respectively. Validation of the predictive model with a holdout sample was not possible because all reference/near reference sites were required for model development. Application of the predictive model to the 53 nonreference and 48 degraded EMAP sites resulted in the classification of 44 (43.6 percent), 15 (14.9 percent), and 42 (41.6 percent) stream sites in Group 1, Group 2, and Group 3, respectively (see Figure 4).

 

Figure 4. Location of Reference and Near Reference Sites (Closed Circles) and Nonreference and Degraded Sites (Open Circles) in Mid-Atlantic Highlands Region by Group. Classification of the latter to one of the three groups was achieved by application of the MDA predictive model. Discriminant functions were based on the following variables: latitude, stream temperature, boulder cover, and stream gradient.

Metrics subsequently were evaluated within each of the three groups and for the MAHA, with all sites combined independent of group membership and the latter serving as a reference point to assess improvement. Secondary examination of metrics revealed that classification was effective in removing the strong latitudinal gradient in all groups, and with respect to Group 1 and Group 3, had accounted for most of the other gradients as well. In Group 2, however, new gradients surfaced that were not visible in the larger data set, including a longitudinal gradient that was stronger than before. These gradients are believed to have surfaced as a result of not having classified sites in Group 2 further. Doing so, however, would have created one more category, reducing the smallest sample size to eight. In spite of lower sample size, a four-group, rather than a three-group classification may have been more effective to remove “noise” observed among Group 2 sites.

In Group 1, values varied significantly between reference and degraded sites for 9 of the 11 metrics. Correct classification for these metrics ranged from 53 percent to 81 percent. In Group 2, only four metrics were significant. The number of correctly classified sites by these metrics was 55 percent to 59 percent. There were seven significant metrics in Group 3; correct classification varied from 21 percent to 86 percent. For the MAHA, all groups combined, 10 of the 11 metrics were significant; correct classification ranged from 55 percent to 78 percent. These results indicate that metric performance varied widely within and among groups. The average group classification efficiency ranged from 56 percent to 66 percent. Group 2 had the lowest.

Average classification efficiency for eight IBIs by group ranged 62 percent to 68 percent, suggesting that collectively, classification efficiency was mediocre at best. Individually, however, IBI performance ranged considerably within and among groups, that is, the same IBI that performed reasonably well in one group (> 70% correctly classified), fared very poorly in another (e.g., IBIs 1- 4). Classification efficiency for Group 2 was consistently poor, and never exceeded 70 percent for the combined classification efficiency (see Table 2).

Table 2. Classification Efficiency for Eight IBIs by Group and for the MAHA (All Sites Combined). Classification efficiency was calculated as the proportion of sites correctly classified as reference, nonreference, and degraded (scores > 50 % of reference for reference sites, > 25 % of reference for nonreference sites, < 25 % of reference for degraded sites). Within each group, the percent of correctly classified sites is given for all three stream types (Cmb), degraded (Deg), and nonreference (Nrf). The number of observations per column is in parenthesis. Except for means, only percentages greater than 60 percent are shown. Refer to Table 20 for the key to the metric identifier. Cells in bold identify the top two IBIs per group.

IBI Name	Identifier for Metcis in IBI					Percent of sites correctly classified
		Group 1			Group 2			Group 3			MAHA
		Cmb (53)	Deg (23)	Nrf (21)	Cmb (31)	Deg (6)	Nrs (9)	Cmb (51)	Deg (19)	Drf (23)	Cmb (135)	Deg (48)	Nrf (53)
"SS IBI"	A E F I	62	81		68	67	78	63	74		62	81
1	D E F I	62	91				67	65	68	65	63	81
2	D E F	66	91		65	67		71	74	74	69	77
3	E F I H	64	87		65	83	6	65	63	70	64	81
4	E F I	68	83		65	67	78	69	63	78	69	81	62
5	A C G	81	83	81		67		65	89			75
6	A C K J	70	61	86				63	89		64	75
7	A K	74	74	81	61		78	76	89	70	76	83	75
Man		68	82	58	62	63	63	67	76	62	66	79	57

Based on the results with the MAHA data set, benefits to assigning test sites to a group before applying an IBI will vary depending on group membership. Poor classification efficiency experienced with Group 2 may be resolved by further classification of this group, an approach that will require a larger data set. Regardless of the approach taken, considerable evidence presented here suggests that strong natural gradients are affecting potentially useful metrics. Unless poor separation of reference and degraded sites are corrected by other means, future improvement of a regional index based on stream salamanders may most likely depend on subdivision of the MAHA into more homogenous geographic regions.

Part II–Volunteer Study

There is a widespread network of citizen volunteer monitoring groups throughout the United States. Equally impressive is the number of organizations dedicated to the support and education of volunteer groups. Streams and rivers are by far the most intensively monitored aquatic habitat.

Recently, there has been substantial interest in developing and testing IBIs based on stream salamanders; a relatively ubiquitous, abundant, and fairly easy to sample taxa in the northern Appalachians. In consideration of the above, the prospect of involving volunteers for sampling salamanders appears very attractive. The objectives of the study were to evaluate the level of proficiency attained by volunteers to sample, process, and most importantly, identify Pennsylvania stream salamanders after training.

The study entailed training and testing volunteers. Volunteers were tested in the classroom and in the field. The classroom provided a controlled environment for testing volunteer identification skills with all seven species and their lifestages. The field task required sampling by the plot method. It was intended to evaluate volunteer identification skills in the field and plot sampling proficiency. Vouchers were used to confirm the identity of a portion of the salamanders sampled by volunteers.

Efforts to recruit individuals to assist with the volunteer phase of the SPAR project (a component of Phase II) took place during March-May 2002. Recruitment efforts included public speaking, electronic publications, creation of Web pages on a Pennsylvania State University Web site, and telephone and e-mail correspondence.

To register as volunteers, respondents interested in participating were required to complete a volunteer application form linked to the Web site. Respondents had to complete 15 questions, 9 of which were aimed at gauging their pretraining exposure and knowledge of stream salamanders. Volunteer training locations were established at University Park, PA, and in the vicinity of Harrisburg, PA. The locations were selected to facilitate travel for volunteers. A total of five 8-hour training sessions were offered on the last week of June 2002. The syllabus consisted of the following components: a pretraining test (30 minutes), an introductory lecture on headwater assessments, salamander identification training aided by 94 slides, a "practicum"—a period in which volunteers examined live Pennsylvania stream Plethodontids, a posttraining test (30 minutes), and discussion of the training manual and field sampling techniques (see Figures 5 and 6).

The pretraining and posttraining tests were identical. The use of these tests was intended to measure how much the volunteers learned during the training. It required volunteers to identify all Pennsylvania stream salamanders and their lifestages. This information also identified training weaknesses. Volunteers were trained and tested only with live specimens. The tests required volunteer groups to identify specimens to species when one or more live animals were presented.

 

Figure 5. Box Plots Summarizing Volunteer Test Scores. (A) Comparison of pretraining to posttraining test scores for the 36 volunteer groups. (B) Effect of group size on posttraining test scores. (C) Pre- and posttraining test scores across the five training sessions. All box plots depict 25th and 75th quartile ranges and medians. The black dots above the range bars depict outliers.

 

Figure 6. Bar Graph Illustrating Pretraining and Posttraining Performance by Test Question. The abscissa reflects the percentage of volunteer groups responding correctly to each question.

Volunteers were asked to sample a stream location of their choice. This posttraining activity was designed to evaluate successful completion of SPAR sampling by volunteer groups in the field. Volunteers were equipped with a sampling kit to assist with sampling, processing, and shipping of vouchers and completed field data forms.

By June 15, 82 individuals representing 56 volunteer groups applied to participate. Training was attended by 65 individuals representing 41 volunteer groups. Of the 64 trainees, 70 percent were biologists and 84 percent had searched for amphibians in the past. All applicants were entered in the study, regardless of their level of experience with amphibians.

Pretraining scores for 38 of the 41 volunteer groups attending the training averaged 49 percent and ranged from 8 percent to 83 percent, a range of 76 percent. Volunteers with prior amphibian knowledge scored higher on average on the pretraining test than volunteers that did not.

Posttraining tests for 39 volunteer groups averaged 81 percent and ranged from 54 percent to 100 percent, a range of 46 percent. This outcome can be interpreted as an improvement in the average test scores of 32 percent and a reduction in the range of scores of 29 percent. The training thus had the dual effect of increasing test scores on average and reducing their variability. Both of these effects were significant. Test scores did not vary significantly among training sessions before training, but varied significantly following the training.

As might be expected, some salamander species and lifestages were more difficult to identify than others. Volunteer proficiency varied by salamander species and lifestage even after the training. Identification of Desmognathus to species was the most challenging, especially when the northern dusky (Desmognathus f. fuscus) and the mountain dusky (Desmognathus ochrophaeus) were in the same set. It should be noted, however, that although many groups failed to make the correct identification to species, most had correctly identified Desmognat hus specimens to genus.

The field sampling task was completed by 23 (56 percent) volunteer groups. Sampling by volunteers occurred from July 6 to October 6, 2002, and took place in 15 Pennsylvania counties. Two sites were located in Maryland. A total of 52 individuals, which included non-SPAR trainees, participated in the sampling effort.

Based on the completed field forms, sampling by volunteers resulted in the capture and processing of 612 salamanders, of which 461 (75 percent) were recorded as larval or transforming (gill stubs), and 151 (25 percent) as terrestrial. The median salamander abundance at volunteer sites (number of salamanders in 3 plots) was 13 and ranged from 1 to 105.

Of the 612 salamanders recorded, 126 (21 percent) were processed and returned as vouchers. Examination of vouchers revealed 89 (71 percent) to be larvae or metamorphs. The remaining 37 (29 percent) were nonlarval. The proportion of larval to nonlarval for the vouchers was very close to that reported for animals processed by volunteers in the field.

Table 3. Summary of Incorrect Taxonomic Identification for 28 Vouchers. A total of 126 vouchers submitted by 23 volunteer groups were examined. Columns from left to right identify: actual identity of vouchers, voucher identity by volunteer, the number of vouchers identified, and the number of volunteer groups involved. Actual lifestage of the vouchers are noted as gilled (g), stubbed (s), and not gilled (ng). The total number of vouchers by lifestage is given in the first column. Vouchers were incorrectly identified by 11 groups. Of these groups, three incorrectly identified vouchers twice (different species).

 

Actual Identity of Vouchers	Volunteer Identification	No. of Vouchers			No of Groups
		Not gilled	Stubbed	Gilled
Mountain dusky n = 152[15 ng]  Groups = 6	N dusky	3			1
N. Dusky n = 13[12ng] Groups = 10	Mountain dusky	1			1
Appalachinn seal n = 3[3ng] Groups = 2	All correctly identified
N. two-lined n = 79[7 ng. 143, 58g] Groups = 22	N. Dusky			16	5
	N. Spring			2	1
	Total			18	6
N. spring n = 3[3 g] Groups = 3	N. red			1	1
	N. two-lined			1	1
	Total			2	2
N. red n = 12 [1 g, 11g] Gropus = 8	N. Spring			2	2
	N. two-lined			1	1
	Total			3	3
redblack N = 1[1 ng] Groups = 1	Mountain duaky	1			1
Total n = 126 [38 ng. 15 g, 73 g] Groups = 23		5		23	11 groups (48%); 3 (13%) groups twice

Laboratory examination of vouchers confirmed the presence of seven species. Of the 126 vouchers examined, 28 (22 percent) were incorrectly identified. Most of these misidentifications were at genus (n = 24) rather than at species (n = 4) level. All of the former were larvae, whereas incorrectly identified species were nonlarvae. Results show that larval Eurycea bislineata were commonly confused with larval D. fuscus, a recurring error that accounted for 16 (67 percent) of the 28 misidentifications. Lifestage was incorrectly identified for 16 (13 percent) specimens.

These results indicate that volunteer training was beneficial, but the level of proficiency attained under these favorable testing conditions varied among volunteer groups tested and depended on the salamander species presented. Future volunteer training efforts may benefit from further instruction and greater focus on the more difficult to identify species and lifestages identified in this study. Better methods or tools for discrimination are needed. Not all trainees completed the sampling task. Those that did appeared to do so satisfactorily, albeit with an "effectiveness" that did not appear comparable to SPAR project crews.

The results suggest that collection of stream salamander data by minimally trained volunteer crews at this level of detail may be most fruitful and reliable when such efforts are conducted in concert with appropriately designed quality assessment/quality control programs that allow confirmation of species identity by whatever methods available. The tasks and expertise demanded in this study, however, may not be necessary depending on the goals and methods of a future proposed sampling/monitoring program.

Journal Articles:

No journal articles submitted with this report: View all 7 publications for this project

Supplemental Keywords:

stream salamander, stream, environmental assessment, sentinel species, ecosystem protection, environmental exposure and risk, chemical mixtures, ecological effects, human health, ecological indicators, ecosystem indicators, ecology, ecosystem assessment, environmental chemistry, hydrology, microbiology, exploratory research environmental biology, Mid-Atlantic Highlands area, MAHA, amphibians, aquatic biota, aquatic ecosystems, ecological exposure, forested headwater ecosystems, salamander population, stream ecosystems, stream plethodontid assemblage response, SPAR., RFA, Ecosystem Protection/Environmental Exposure & Risk, Scientific Discipline, Ecological Indicators, Ecological Risk Assessment, Ecology, Environmental Chemistry, Ecosystem/Assessment/Indicators, Microbiology, Ecological Effects - Environmental Exposure & Risk, Hydrology, stream ecosystems, aquatic biota , salamander population, amphibians, stream plethodontid assemblage response (SPAR), ecosystem indicators, MAHA, ecological exposure, aquatic ecosystems, forested headwater ecosystems

Progress and Final Reports:

Original Abstract

2000 Progress Report

2001 Progress Report