Printer Friendly

Model distribution of silver chub (Macrhybopsis storeriana) in Western Lake Erie.

INTRODUCTION

The Silver Chub (Macrhybopsis storeriana Kirtland, 1845), a once common Lake Erie species, is now rare. It is listed as Endangered in New York, Ohio, and Ontario. The bulk of its distribution is within the Mississippi River system (Page and Burr, 2011; FishBase, 2013; DCNR, 2013). However, western Lake Erie may hold the last remaining substantial large lake population of this species; its population in Lake Erie crashed in the 1950s and 1960s (Parker et al., 1987; NYDEC, 2013; COSEWIC, 2013). Little is known of Silver Chub life history or habitat requirements, but it is a relatively large minnow (<230 mm, 170 g) and was historically a component of the forage fish assemblage ([greater than or equal to] 49/ha in Lake Erie) generally found in waters <10 m deep (to maximum of about 20 m) (Kinney, 1954; Smith, 1985; Coker et al., 2001; ODNR, 2013; COSEWIC, 2013). Silver Chub is associated with a wide variety of bottom types but is generally found in open water and not associated with submerged aquatic vegetation (SAV). It prefers warm waters and among the Great Lakes temperature may limit this species to Lake Erie and Lake St. Clair (Kinney, 1954), although there is one record of Silver Chub in Lower Lake Huron (COSEWIC, 2013).

Silver Chub was historically a substantial component of the Lake Erie aquatic community and may play a larger role if its population recovers. Knowledge of Silver Chub habitat distribution within western Lake Erie and associated broad-scale habitat conditions may help conserve this species by identifying critical habitats that could be protected (Boyko and Staton, 2010). Species distribution models (SDM) capture essential elements of a species-habitat relationship and predict the species distribution in all areas with suitable habitats, usually displayed on a map. These models have been widely used in Conservation Biology and successfully applied to identify habitats capable of supporting many different species. However, within-lake SDMs have rarely been used to map predicted aquatic species distributions (Bergstedt et al., 2003; Douglas et al., 2009; McKenna and Castiglione, 2010; Simonin et al., 2012). This work reports on the first Silver Chub distribution model and associated suitable habitat conditions. Our objectives were to: (1) train artificial neural networks (NN) with standardized trawl and electrofishing data and matching habitat conditions to predict Silver Chub abundance, (2) apply those models to project the potential distribution of this species throughout western Lake Erie, and (3) examine the habitat conditions associated with various qualities of Silver Chub habitat.

METHODS

Electrofishing-collected samples of Silver Chub within 43 sampling sites selected to represent nearshore habitats in shallow coastal waters of Ohio (June-October 1982-2002) (41[degrees]23'48"N-41[degrees]41'24"N, 82[degrees]32'55"W-83[degrees]28'33"W) were provided by the Ohio Environmental Protection Agency (EPA) (Ohio Environmental Protection Agency, 2003; Thoma, 1999); these collections generally occurred in <5 m of water. Trawl-collected samples of Silver Chub within 62 fisheries sampling areas established by the Ohio Department of Natural Resources (DNR) throughout U.S. waters of western Lake Erie (41[degrees]25'55"N41[degrees]54'35"N, 82[degrees]33'59"W-83[degrees]25'37"W) were available for the period 1987-2002 (May-October) (Ohio Department of Natural Resources, 2003); these collections generally occurred in >5 m of water. Silver Chub catches from both of these data sets were standardized to number/ ~1000 [m.sup.2] but modeled separately because of the different collection methods (McKenna and Castiglione, 2010). Standardized abundances were classified into broad abundance categories on a log scale using classes of 0, 1, 2-10, 11-100, >100. Each collection was matched with a suite of broad-scale habitat conditions (representing the same general time period as the fish collections) that were available from the Great Lakes Regional Aquatic Gap Analysis Project for each 270 m X 270 m spatial cell within the western basin of Lake Erie (41[degrees]24'56"N- 42[degrees]2'18"N, 82[degrees]27'58"W- 83[degrees]30'27"W) (Table 1, Appendix 1). The observed species-habitat data sets were used to train simple backpropagadon neural networks with a single layer of hidden neurons, following the methods of McKenna and Castiglione (2010) (NeuroShell 2.0 software, Wards Systems Group, Inc., Frederick, Maryland). We used NNs to develop each model because they typically produce more effective predictive models than many classical modeling techniques, particularly when relationships are multivariate and nonlinear; NNs are not dependent upon assumptions about the underlying response model and specified error structure. (Hertz et al., 1991; Olden and Jackson, 2001; McKenna et al., 2006). Habitat variable values were standardized and the influence of each on predictions of Silver Chub abundance class was determined by tracing the changes in weightings through the NN (Olden et al., 2004; McKenna, 2005). Model performance was measured with adjusted coefficient of determination ([R.sup.2]) and Mean Squared Error (MSE) of raw model predictions and omission (predicting absence when the species is present) and commission (predicting presence when the species was not observed) error rates of presence and absence. Cohen's Kappa (K) is reported to provide a measure of model performance relative to chance predictions (K > 0.2 is fair, K > 0.6 is substantial, Landis and Koch, 1977) and was applied to overall presence-absence predictions and each abundance class (Olden et al., 2002). The NN models were applied in a Geographic Information System [ArcGIS 10 (ESRI, Redlands, California, U.S.A.)] and used to predict the abundance class of Silver Chub within each spatial cell of western Lake Erie. The EPA-based NN was used to predict Silver Chub abundance in waters [less than or equal to] 5 m deep and the DNR-based NN was used for predictions in deeper waters. Zonal statistics were used to calculate averages of associated habitat variables within areas predicted to be optimal (>10 Silver Chub/1000 [m.sup.2]) or marginal (<10 Silver Chub/1000 [m.sup.2]), and where Silver Chub was expected to be absent. Signatures of habitat conditions associated with each of these habitat qualities are presented as bar charts of z-scores to show preferred conditions, relative to average conditions throughout the study area.

RESULTS

The EPA and DNR data sets represented the spectrum of habitats in western Lake Erie and observed Silver Chub abundances ranged from zero to >100/ha sampled. The highest abundances were in offshore areas (absent from 6% of DNR samples) and Silver Chub only occurred in five of the shallow, nearshore EPA collections (absent from 88% of EPA samples); most of those were among the islands (Fig. 1).

The DNR NN used 11 hidden neurons and six input variables (most heavily weighted to least): cosine of direction to nearest delta wetland [cdeltdi], mean surface water temperature [xtemp], distance to nearest open wetland [opendis], water depth [depth2], SAV covering [greater than or equal to] 50% of bottom [SAV], and sinuosity of the nearest shoreline [sinuous] (Table 1). The EPA NN used ten hidden neurons and seven input variables (most heavily weighted to least): nearest shoreline geomorphic type [geomorf], distance to nearest protected wetland [protdis], cosine of direction to nearest open wetland [copendi], coefficient of variation of surface water temperature [tempcv], distance to nearest large river mouth [rivdist], distance to nearest open wetland [opendis], and coastal reinforcement condition of nearest shoreline [protect].

Neural network models generally performed well and explained >86% of variation in each data set (Table I). Mean Squared Error was high for the DNR model because of the wide range in abundances of Silver Chub. The omission rate was 0% for the EPA model and 19% for the DNR model as presence-absence (Table 2a, b). However, omission rate for the highest abundance class predictions was 0% for the DNR-based model (Table 2c). Cohen's Kappa indicated that EPA-based model predictions were strong (K = 0.73), but that on a presence-absence basis the DNR model was not much better than chance alone (K = 0.03). However, when used to measure performance of each abundance class (only DNR data included abundances exceeding 1), K shows that model predictions for the most abundant class (>100) were strong (K = 0.77), those for the moderate abundance class (11- 100) were fair (K = 0.41), and those for the lowest abundance class (2-10) were weak (K = 0.13); K values were greater than the associated prevalence for the two highest abundance classes (prevalence, >100: 17.7%, 11-100: 38.7%, 2-10: 37.1%, overall DNR model prevalence: 93.5%); the prevalence of Silver Chub in the EPA data set was 11.6%. Thirty-five percent of the omission errors in the 2-10 class occurred at areas in waters <5 m deep.

The map of predicted distributions is a composite of the two model's predictions, with the EPA model applied in waters [less than or equal to] 5 m deep and the DNR model applied in waters >5 m deep (Fig. 1). The models predict optimal habitat throughout much of the open waters of the central portion of the western basin and marginal habitat in patches in shallower areas near shore, particularly along the southern and western coasts and around the islands. Silver Chub was predicted to be absent from patches mixed among the marginal habitats of shallow nearshore areas.

Signatures of relative habitat conditions reflect the prediction map and indicate that optimal Silver Chub habitat occurs in relatively deep water (>5 m), far from coastal wetlands and major rivers, where SAV is largely absent and water temperature variability is relatively low (Fig. 2). Marginal habitat was associated with shallow waters close to open and protected coastal wetlands and major rivers with an abundance of SAV, along relatively sinuous shorelines with highly variable water temperature. Silver Chub were absent from areas with conditions that were generally moderate between that of the two other abundance classes, but artificial shorelines and flat wetland or bedrock geomorphology and the lowest mean water temperatures were emphasized in these areas.

DISCUSSION

Observations and model projections suggest changes in habitat and ecological function in the Great Lakes has probably contributed to the decline of the Lake Erie population of Silver Chub. This once common species is now listed as Endangered and believed to be extirpated from New York waters (Greeley, 1929; Kinney, 1954; NY DEC, 2013), but fairly high numbers (>100/ha) remain in parts of western Lake Erie (Ohio Department of Natural Resources, 2003). Our NN models were generally effective at predicting habitat capable of supporting various abundance classes of Silver Chub within western Lake Erie (high [R.sup.2] and K values), particularly the highest (mostly in open water areas) and lowest (nearshore areas) abundance classes. Additional data on fish occurrence and abundance and associated habitat conditions, particularly in Canadian waters (which were not represented in our data) and recent collections near the islands (where optimal habitat is in close proximity to shallow habitats), would help to validate and refine these models.

Results indicate that a range of habitat quality exists for Silver Chub in the western Basin of Lake Erie and large areas of open water had the highest quality. Other areas may have the potential to support lesser abundances of the species in shallower nearshore areas. The NN models emphasized distance and position of habitat relative to coastal wetlands, water depth, and water temperature and its variability as the most influential variables. The mapped predictions show optimal habitat conditions in the deepest open water areas of western Lake Erie (still generally <10 m deep) and that marginal habitat exists in shallow areas along the coast and around the islands. Silver Chub were not associated with waters adjacent to modified shorelines or those of particularly low relief and were absent where waters were coolest. Predictions generally align with what is known of this species' preference for open waters [less than or equal to] 10 m deep with relatively stable, warm temperatures (Kinney, 1954; Smith, 1985; Parker et al., 1987; COSEWIC, 2013). However, predicted optimal habitat conditions contrast with some recent observations of large catches from areas close to a large river (Detroit River) where SAV growth is rich (Kocovsky, pers. comm.). Seasonal differences may explain the discrepancy, but additional research on the life history and habitat requirements of this species is clearly required to understand threats to the Lake Erie population and factors affecting its abundance and distribution.

Moderate abundance classes were predicted to be supported by habitats in areas of the southern portion of the basin near the islands, but K values indicated that model predictions were only fair for those classes. Many of the prediction errors were associated with trawl collections in waters <5 m deep and it is possible that the trawls were not as effective at collecting Silver Chub as electrofishing when this species was in low abundance. However, known habitat and diet information suggest that Silver Chub is a demersal species and electrofishing may not be effective in deep waters (~2-5 m). While effort was accounted for in the data standardization, no Silver Chub collection efficiencies were available for the gear.

Food availability, food quality, water quality, and seasonality were not explicitly considered by these models but are likely to affect the distribution and abundance of this species. Silver Chub is known to consume molluscs and relied heavily on Hexagenia spp. mayflies, which declined greatly about the time that Silver Chub also decreased (Kinney, 1954; Coker et al., 2001; DCNR, 2013; NYS DEC, 2013). Recent rebound of Hexigenia spp. populations and invading Driessenid mussels may now supply Silver Chub with its nutritional needs. Our models are focused on the summer time period and indicate that some shallow nearshore areas provide marginal habitat for Silver Chub, at that time. In spring, Silver Chub have been observed moving into shallow waters, which may be a thermotaxic response or could be associated with spawning activity (Kinney, 1954). Shallow nearshore areas are close to pollutant sources from tributaries and water quality may be poor, possibly inhibiting spawning success or making foraging less efficient than in deeper waters. Similarly, hypoxia may restrict recovery of Silver Chub populations, particularly in other parts of Lake Erie where it is a persistent problem (Pothoven et al., 2009). Climate change may favor this species because of its preference for warm temperatures, but associated changes in lake water levels or hypoxic conditions may reduce the amount of appropriate habitat available.

Our model predictions help to identify areas where Silver Chub may still be common and areas that might support substantial populations if conditions were improved. The model predictions represent the potential for habitat units to support a given abundance of this species. Acute anthropogenic factors have been filtered out and where fewer than expected Silver Chub are found, habitat conditions may be degraded below that habitat's potential. Such sites may be candidates for habitat restoration. Similarly, validated optimal habitats may be candidates for protected areas, dj Our modeling method is applicable to any species where georeferenced data are available and has been used to model dozens of other species (McKenna and Castiglione, 2010). Species assemblages may also be modeled in a similar way (McKenna et al, 2013). These models may be re-applied to test the effects of habitat changes by altering input values. It is clear that Silver Chub distribution and abundance is strongly affected by habitat conditions at a variety of spatial scales (either directly by variables used here or others for which these are surrogates), as has been shown for other species (McKenna et al., 2006, 2012, for example) and should be an important consideration in conservation strategies.

APPENDIX 1.--Summary statistics of habitat variables used for either
the OH DNR or OH EPA neural network models (Habitat codes are defined
in Table 1). Q1 and Q3 are lower and upper quartiles, respectively,
sd is standard deviation. Variables used in a particular model are
indicated by boldface data set name

Variable   Data set    Maximum       Q3         Mean       Median

CDELTDI    DNR#            0.99        0.13       -0.37       -0.64
           EPA             0.72       -0.05       -0.40       -0.57
COPENDI    DNR             1.00        0.48       -0.22       -0.35
           EPA#            1.00        0.75        0.02        0.39
DEPTH      DNR#           12.36        8.90        6.60        7.06
           EPA             9.07        1.23        1.16        0.55
GEOMORF    DNR            16          11           9.58       11
           EPA#           16          15          11.79       11
OPENDIS    DNR#       36,527.59   24,480.71   17,257.64   16,514.27
           EPA#       37,742.53   22,587.93   15,841.34   13,829.06
PROTDIS    DNR        26,144.05   16,209.63   12,475.44   12,922.27
           EPA#       22,447.20   16,772.80     9556.80     6455.87
PROTECT    DNR             4           4           2.98        3
           EPA#            5           4           2.84        3
RTVDIST    DNR        25,996.45   19,282.90   14,616.90   15,236.81
           EPA#       26,553.76   16,422.66     9040.88     6102.27
SAV        DNR#            1           0           0.12        0
           EPA             1           0           0.23        0
SINUOUS    DNR#         2820.19     2092.10     1618.22     1473.95
           EPA        29,138.00     2319.00     3554.79     1477.00
TEMPCV     DNR             9.76        6.82        6.25        5.82
           EPA#           15.02       10.95        9.41        9.29
XTEMP      DNR#           24.84       24.12       23.94       23.99
           EPA            24.30       23.91       23.61       23.64

Variable   Data set     SD         Q1       Minimum

CDELTDI    DNR#          0.62       -0.90     -0.99
           EPA           0.52       -0.91     -1.00
COPENDI    DNR           0.71       -0.92     -1.00
           EPA#          0.81       -0.99     -1.00
DEPTH      DNR#          3.09        3.94      0.67
           EPA           1.76        0.18     <0.5
GEOMORF    DNR           4.05        9         1
           EPA#          3.78       10         3
OPENDIS    DNR#       8929.93   10,623.90   1583.23
           EPA#       9827.50     6393.30   1458.82
PROTDIS    DNR        5911.31     8826.96    305.77
           EPA#       7096.06     4232.31     42.43
PROTECT    DNR           1.13        3         1
           EPA#          1.51        1         1
RTVDIST    DNR        6452.79   10,327.63   1321.25
           EPA#       8405.97     1827.79    384.85
SAV        DNR#          0.30        0         0
           EPA           0.43        0         0
SINUOUS    DNR#        524.17     1100.50   1000.00
           EPA        5994.12     1185.00   1000.00
TEMPCV     DNR           1.11        5.53      4.99
           EPA#          2.20        7.78      5.57
XTEMP      DNR#          0.29       23.76     23.20
           EPA           0.45       23.43     22.13

Note: Variables used in a particular model are indicated with # data
set name.


Acknowledgments.--We are indebted to the Ohio Department of Natural Resources and Ohio Environmental Protection Agency for providing fish collection data and to the Great Lakes Regional Aquatic Gap Analysis Project for standardizing the fish collections and providing habitat data within western Lake Erie. We would like to thank M. Slattery for assistance with GIS mapping. We are also grateful to P. Kocovsky, N. Mandrak, K. McKenna, and other reviewers for their comments and suggestions. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government. This is Contribution 1792 of the U.S. Geological Survey Great Lakes Science Center.

LITERATURE CITED

Bergstedt, R. A., R. L. Argyle, J. G. Seelye, K. T. Scribner, and G. L. Curtis. 2003. In situ setermination of the annual thermal habitat use by lake trout (Salvelinus namaycush) in Lake Huron. J. Great Lakes Res., 29 (Supplement 1):347-361.

Boyko, A. L. and S. K. Staton. 2010. Management plan for the Silver Chub, Macrhybopsis storeriana, in Canada. Species at risk act management plan series. Fisheries and Oceans Canada, Ottawa, vi + 21 pp. http://www.sararegistry.gc.ca/virtual_sara/files/plans/mp_silver_chub_0910_e.pdf . Last visited 25 July 2013.

Coker, G. A., C. B. Portt, and C. K. Minns. 2001. Morphological and ecological characteristics of Canadian freshwater fishes. Can. Manuscr. Rep. Fish. Aquat. Sci., 2554. January 2001.

COSEWIC (Committee on the Status of Endangered Wildlife in Canada). 2013. COSEWIC assessment and status report on the Silver Chub Macrhybopsis storeriana in Canada. Committee on the Status of Endangered Wildlife in Canada, Ottawa, xiii + 34 pp. http://www.sararegistry.gc.ca/ default.asp?lang=En&n=68D8614A-l. Last visited 25 July 2013.

DCNR. 2013. Web page: http://www.outdooralabama.com/fishing/freshwater/fish/other/minnow/ chub/silver. 1/14/2013. Alabama Department of Conservation and Natural Resources.

Douglas, J., T. Hunt, N. Abery, and M. Allen. 2009. Application of GIS modelling to quantify fish habitats in lakes. Lakes Reservoirs: Res. Manage., 14. 2:171-174.

FishBase. 2013. Web page: (http://www.fishbase.org/summary/Macrhybopsis- storeriana.html. 2/12/ 2013). FishBase, A Global Information System on Fishes.

Hertz, J., R. Krough, and R. Palmer. 1991. Introduction to the theory of neural networks. Addison-Wesley, Redwood City, California.

Kinney, E. M. S. 1954. A life history of the silver chub, Hybopsis storeriana (Kirtland), in western Lake Erie with notes on associated species. Ph.D. Thesis, Ohio State University, Columbus, Ohio. Landis, J. R. and G. G. Koch. 1977. The measurement of observer agreement for categorical data. Biometrics., 33:159-174.

McKenna, J. E. Jr. 2005. Application of neural networks to prediction of fish diversity and salmonid production in the Lake Ontario Basin. Trans. Am. Fish. Soc., 134:28-43.

--and C. Castiglione. 2010. Hierarchical multi-scale classification of nearshore aquatic habitats of the Great Lakes: Western Lake Erie./ Great Lakes Lies., 36:757-771.

--, D. M. Carlson, and M. L. Payne-Wynne. 2013. Predicting locations of rare aquatic species' habitat with a combination of species-specific and assemblage-based models. Diversity Dist., 19:503-517.

--, J. E. Ruggirello, and J. H. Johnson. 2012. A landscape-based distribution model for fallfish (Semotilus corporalis) in the Great Lakes drainage of New York./ Great Lakes Res., 38:413-417.

--, R. P. McDonald, C. Castiglione, S. Morrison, K. Kowalski, and D. Passino- Reader. 2006. A broadscale fish-habitat model development process: Genesee Basin, New York, p. 533-554. In: R. M. Hughes, L. Wang, and P. W. Seelbach (eds.). Landscape influences on stream habitats and biological assemblages. Am. Fish. Soc., Symp. 48, Bethesda, Maryland.

NYDEC. 2013. Silver Chub Fact Sheet. Web page: http://www.dec.ny.gov/animals/26010.html. 1/24/ 2013. New York State Department of Environmental Conservation.

ODNR. 2013. Silver Chub web page: http://www.dnr.state.oh.us/Home/species_a_to_z/Species GuideIndex/silverchub/tabid/223. 1/24/2013. Ohio Department of Natural Resources.

Ohio Department of Natural Resources. 2003. Ohio Division of Wildlife, Lake Erie Fisheries 2002, Annual Status Report, Federal Aid in Fish Restoration F-69p, Ohio Department of Natural Resources, Division of Wildlife, Lake Erie Fisheries Unit.

Ohio Environmental Protection Agency. 2003. Biological Monitoring Data: Lake Erie. Ohio Environmental Protection Agency, Available from: Division of Surface Water, 50 West Town Street, Suite 700, Columbus, Ohio.

Olden, J. D. and D. A. Jackson. 2001. Fish-habitat relationships in lakes: gaining predictive and explanatory insight by using artificial neural networks. Trans. Am. Fish. Soc., 130:878-897.

--, --, and P. R. Peres-Neto. 2002. Predictive models of fish species distributions: a note on proper validation and chance predictions. Trans. Am. Fish. Soc., 131:329-336.

Page, L. M. and B. M. Burr. 2011. Peterson field guide to freshwater fishes of North America north of Mexico. 2nd ed. Houghton Mifflin Harcourt, New York. 663 p.

Parker, B., P. McKee, and R. R. Campbell. 1987. Status of the silver chub, Hybopsis storeriana, in Canada. Can. Field-Nat., 101:190-194.

Pothoven, S. A., H. A. Vanderploeg, S. A. Ludsin, T. O. Hook, and S. B. Brandt. 2009. Feeding ecology of emerald shiners and rainbow smelt in central Lake Erie. J. Great Lakes Res., 35:190-198.

Simonin, P. W., D. L. Parrish, L. G. Rudstam, P. J. Sullivan, and B. Pientka. 2012. Native rainbow smelt and nonnative alewife distribution related to temperature and light gradients in Lake Champlain. J. Great Lakes Res., 38:115-122.

Smith, C. L. 1985. The inland fishes of New York State. New York State Department of Environmental Conservation, Albany, New York. 522 p.

Thoma, R. F. 1999. Biological monitoring and an index of biotic integrity for Lake Erie's nearshore waters, p. 418-461. In: T. P. Simon (ed.). Assessing the sustainability and biological integrity of water resources using fish communities. CRC Press, Boca Raton, Florida.

SUBMITTED 25 MARCH 2013

ACCEPTED 30 SEPTEMBER 2013

JAMES E. McKENNA, Jr. (1)

Tunison Laboratory of Aquatic Science, U.S. Geological Survey, Great Lakes Science Center, 3075 Grade Road, Cortland, New York 13045

AND

CHRIS CASTIGLIONE (2) V. S. Fish and Wildlife Service, Lower Great Lakes Fish and Wildlife Conservation Office, 1101 Casey Road, Basom, New York 14013

(1) Corresponding author: Telephone: 607-753-9391 x7521; e-mail: jemckenna@usgs.gov

(2) Telephone: 585-948-5445; e-mail: chris_castiglione@fws.gov

TABLE 1.--Summary of neural network models' weights for each data
set. Adjusted [R.sup.2] and Mean Squared Error are shown below each
data set name

                                                                 Mean
Data set                          Variable [code]               weight

Ohio DNR              cosine of direction to nearest delta       3.557
[R.sup.2] = 0.8618,     wetland [cdeltdi]
  MSE = 4596          mean surface water temperature [xtemp]     3.511
                      sinuosity of the nearest shoreline        -0.775
                        [sinuous]
                      SAV covering [greater than or equal       -0.826
                        to] 50% of bottom [sav]
                      water depth [depth2]                      -1.743
                      distance to nearest open wetland          -3.148
                        [opendis]

Ohio EPA              nearest shoreline geomorphic type          7.697
[R.sup.2] = 0.8645,     [geomorf]
  MSE = 0.04          cosine of direction to nearest open        6.228
                        wetland [copendi]
                      coastal reinforcement condition of         0.424
                        nearest shoreline [protect]
                      distance to nearest open wetland           0.055
                        [opendis]
                      distance to nearest large river mouth     -2.729
                        [rivdist]
                      coefficient of variation of surface       -5.431
                        water temperature [tempcv]
                      distance to nearest protected wetland     -7.346
                        [protdis]

TABLE 2.--Correct predictions and omission and commission errors.
Omissions are in bold and commissions are in italics. Cohen's Kappa
(K) and prevalence are also shown

a. Ohio EPA model (K = 0.73, prevalence = 11.6%)

                               Observed

            Abundance class    0      1

Predicted         0#           35     0
                  1#          3##     5

b. Ohio DNR Presence-absence model (K = 0.03, prevalence = 93.5%)

                               Observed

            Abundance class    0      >0

Predicted         0#           1     11#
                  >0#         3##     47

c. Ohio DNR model: Maximum abundance class (K = 0.76, prevalence =
17.7%)

                               Observed

            Abundance Class    0     >100

Predicted         0#           46     0#
                 >100#        5##    11#

d. Ohio DNR model: Moderate abundance class (K = 0.42, prevalence =
38.7%)

                               Observed

            Abundance Class    0     >10

Predicted         0#           26     6#
                11-100#       12##   18#

e. Ohio DNR model: Marginal abundance class (K = 0.13, prevalence =
37.1%)

                               Observed

            Abundance Class    0     2-10

Predicted         0#           38    20#
                 2-10#        1##     3

Note: Omissions are indicated with # and commissions are indicated
with ##.
COPYRIGHT 2014 University of Notre Dame, Department of Biological Sciences
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:McKenna, James E., Jr.; Castiglione, Chris
Publication:The American Midland Naturalist
Article Type:Author abstract
Geographic Code:1U2NY
Date:Feb 1, 2014
Words:4407
Previous Article:Current and historical extent of phenotypic variation in the tufted and black-crested titmouse (Paridae) hybrid zone in the southern great plains.
Next Article:Distribution and habitat correlates of clinch dace (Chrosomus sp. cf. saylori) in the upper Clinch River watershed.
Topics:

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |