Evaluation of Predicted Fish Distribution Models for Rare Fish Species in South Dakota

Abstract

Predictive fish distribution models have utility in planning conservation measures for rare fish species. However, species rarity creates sampling and modeling difficulties that require an understanding of model accuracy. We evaluated existing distribution models for 10 rare fishes based on 2,026 community fish samples and associated riverine habitat. Our fieldwork provided an independent fish species inventory for 143 sample sites. This inventory was used to quantify species detectability for use as a weighting factor with which to correct false-negative modeling errors. Presence/absence data were compared with predictions to evaluate model accuracy as determined by Cohen's kappa and correct classification rates. Detection probabilities were generally small but ranged from 0 to 0.68. The models predicted species occurrence with relatively high success (average kappa ¼ 45.6% and average correct classification rate ¼ 74.1%). The habitat variables for predicting species occurrence varied among species; however, stream size and streamflow were the most influential. All distribution models had adequate predictive abilities and improved our understanding of fish distributions and the factors determining those distributions. Model accuracy statistics can provide managers with a measure of confidence when they are directing conservation activities. a Species with one asterisk are listed as threatened in South Dakota, species with two asterisks as endangered (SDGFP 2007). 1262HAYER ET AL.

Habitat Associations of Fish Species of Greatest Conservation Need at Multiple Spatial Scales in Wadeable Iowa Streams

“…Landis and Koch (1977) arbitrarily characterized κ values of less than 0.40 as indicating poor to fair model performance and κ values greater than 0.40 as indicating moderate to nearperfect model performance. Cohen's κ is commonly used to evaluate fish species presence-absence models (e.g., Olden and Jackson 2001;Rashleigh et al 2005;Hayer et al 2008) and is also used in the medical and remote sensing fields (Congalton 1991;Manel et al 2001). Kappa has limitations and may be overly sensitive to species prevalence (McPherson et al 2004;Vaughan and Omerod 2005); however, Manel et al (2001) found that κ was a robust indicator of model performance and was negligibly influenced by species prevalence.…”

Fish Species of Greatest Conservation Need in Wadeable Iowa Streams: Current Status and Effectiveness of Aquatic Gap Program Distribution Models

“…Species distribution models are an important tool for addressing many natural resources issues and are commonly used to describe species-habitat relationships, identify suitable habitats for species reintroductions, prioritize areas for monitoring and conservation efforts, and forecast the effects of land use changes and habitat alterations Guisan and Thuiller 2005;Lyons et al 2010). As habitat loss and degradation continue to threaten fish biodiversity in North America (Miller et al 1989;Richter et al 1997;Jelks et al 2008), species distribution models are playing an increasingly important role in fisheries conservation and management (Wall et al 2004;Dauwalter and Rahel 2008;Hayer et al 2008). However, modeling the distributions of fish species is challenging because their occurrences are influenced by a combination of abiotic and biotic factors acting at multiple spatial scales (Poff 1997;Marsh-Matthews and Matthews 2000;Jackson et al 2001).…”

Reach and catchment‐scale characteristics are relatively uninfluential in explaining the occurrence of stream fish species

“…() found that four samples spread across seasons are sometimes needed to detect all of the fishes that occur at a particular site in prairie streams. Inclusion of temporal variability in fish assemblages and habitat conditions or detection probability in species‐habitat models has been suggested by other authors to improve accuracy (Wiley et al ., ; Hayer et al ., ; Steen et al ., ; Falke et al ., ) and may have improved the number of significant reach and catchment‐scale models in this study. Matthews () in a study of fish communities in riffle habitats of the Roanoke River, Virginia, U.S.A., and Mullen et al .…”