SEARCH

SEARCH BY CITATION

Keywords:

  • classification trees;
  • climate change;
  • geographic information systems;
  • habitat;
  • water temperature

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References

Summer air and stream water temperatures are expected to rise in the state of Wisconsin, U.S.A., over the next 50 years. To assess potential climate warming effects on stream fishes, predictive models were developed for 50 common fish species using classification-tree analysis of 69 environmental variables in a geographic information system. Model accuracy was 56·0–93·5% in validation tests. Models were applied to all 86 898 km of stream in the state under four different climate scenarios: current conditions, limited climate warming (summer air temperatures increase 1° C and water 0·8° C), moderate warming (air 3° C and water 2·4° C) and major warming (air 5° C and water 4° C). With climate warming, 23 fishes were predicted to decline in distribution (three to extirpation under the major warming scenario), 23 to increase and four to have no change. Overall, declining species lost substantially more stream length than increasing species gained. All three cold-water and 16 cool-water fishes and four of 31 warm-water fishes were predicted to decline, four warm-water fishes to remain the same and 23 warm-water fishes to increase in distribution. Species changes were predicted to be most dramatic in small streams in northern Wisconsin that currently have cold to cool summer water temperatures and are dominated by cold-water and cool-water fishes, and least in larger and warmer streams and rivers in southern Wisconsin that are currently dominated by warm-water fishes. Results of this study suggest that even small increases in summer air and water temperatures owing to climate warming will have major effects on the distribution of stream fishes in Wisconsin.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References

Water temperature is a key factor determining fish distribution and assemblage composition (Magnuson et al., 1979). In north-temperate streams, summer maximum water temperatures are particularly important, and a reach with a relatively cold summer maximum (cold water) will have an almost completely different fish fauna than a reach with a relatively warm maximum (warm water) (Lyons et al., 2009). Changes in maximum water temperature of only a few degrees can result in major shifts in the occurrence, abundance, survival, growth, size distribution and reproduction of many species (Magnuson et al., 1997; McGinn, 2002; Hughes et al., 2006; Ficke et al., 2007).

Summer maximum water temperature in streams is determined in large part by air temperature (Stefan & Preud’homme, 1993; Erickson & Stefan, 2000). Climate change models for the Laurentian Great Lakes region of east-central North America project increases in annual air temperatures of 1–8° C by the year 2050 (Magnuson et al., 1997), and if these models are correct, then stream summer maximum water temperatures will also increase (Pilgrim et al., 1998; Mohseni et al., 1999), with major consequences for stream fishes (Eaton & Scheller, 1996; Mohseni et al., 2003). The exact relation between air temperature and stream water temperature, however, is complex and influenced by groundwater contributions to streams, stream channel form, riparian vegetation and shading, precipitation and rates of evaporation, such that the water temperatures of individual stream reaches respond differently to a given rise in summer air temperature (Meisner et al., 1988; Erickson & Stefan, 2000; Mohseni et al., 2002; Chu et al., 2008). Anthropogenic factors further influence the air temperature–water temperature relation, and maintenance or rehabilitation of natural vegetation in the stream catchment, particularly the riparian area, can mitigate increases in water temperature and changes in fish assemblages associated with an increase in air temperature (Peterson & Kwak, 1999; Whitledge et al., 2006; Marshall et al., 2008).

Numerous studies have considered the potential effects of climate change on the distribution and abundance of stream fishes in North America (Eaton & Scheller, 1996; McGinn, 2002; Mohseni et al., 2003). These studies have yielded important insights, but most have been relatively coarse in scale, focusing either on broad landscapes where study sites represent only a small fraction of the total stream habitat available (Eaton & Scheller, 1996; Mohseni et al., 2003) or on general patterns within relatively large spatial units such as catchments >500 km2 (Jackson & Mandrak, 2002; Chu et al., 2005, 2008) or thermal isopleths in relation to elevation or latitude (Johnson & Evans, 1990; Meisner, 1990a; Shuter & Post, 1990; Keleher & Rahel, 1996; Rahel et al., 1996; Magnuson et al., 1997; Rahel, 2002; Shuter et al., 2002; Flebbe et al., 2006). Relatively few studies have emphasized the small-scale variation in thermal conditions that may exist within and among individual streams in a region or simultaneously considered the role of other non-thermal habitat factors in determining the distribution of stream fishes, and these studies have tended to cover relatively small geographic areas (Meisner, 1990b; Jager et al., 1999; Battin et al., 2007; Nelson & Palmer, 2007; Nelson et al., 2009; Williams et al., 2009; Steen et al., 2010).

Understanding and accounting for both small-scale variation in stream thermal conditions and the influence of non-thermal habitat factors are essential to better predict responses of stream fishes to climate change in the Laurentian Great Lakes region of North America. Here, stream summer maximum water temperatures are naturally highly heterogeneous across the landscape. Cold-water, cool-water and warm-water reaches occur in close proximity and under the same climate conditions because of local variation in geology and groundwater contributions to streams (Lyons, 1996; Wiley et al., 1997; Zorn et al., 2002; Wehrly et al., 2003; Stanfield et al., 2006; Lyons et al., 2009; Diebel et al., 2010). In this region, coarse-scale analyses may miss local features that could allow cold-water habitats and associated fish species to persist in the face of rising air temperatures and consequently overestimate habitat losses for cold-water fishes and gains for warm-water fishes. Fine-scale analyses that look solely at air and water temperatures, however, will be incomplete, and must also consider other important limiting habitat factors such as stream flow, gradient, valley and channel form, geology and land cover to fully understand patterns of species distribution and responses to climate change (Lyons, 1996; Wang et al., 1997, 2000; Diebel et al., 2010; Steen et al., 2010).

Recent advances in database development and management, geographic information systems (GIS) capabilities and statistical analyses have greatly increased the ability to consider small-scale variation in stream conditions across broad landscapes (Fisher & Rahel, 2004; Hughes et al., 2006). It is now possible to develop and apply multiple predictive models of fish occurrence and abundance that encompass hundreds of thousands of discrete stream reaches ranging in length from tens to thousands of metres (Lyons et al., 2009). Such models allow for much more precise predictions of the response of stream fishes to climate change and rising water temperatures. In this paper, models to predict the amount and distribution of suitable stream habitat in relation to catchment, riparian and channel conditions are developed for 50 stream and river fish species in the state of Wisconsin, U.S.A; located in the western portion of the Laurentian Great Lakes region. These models are applied to thousands of discrete reaches encompassing all the streams and rivers in the state under four different thermal conditions in this initial assessment of how climate warming over the next 50 years might influence stream fish distributions.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References

The modelling approach presented here had three main steps. First, data on fishes and environmental data from a relatively small number of well-studied stream sites were compiled and used to develop accurate predictive models of fish occurrence. These empirical models incorporated particular environmental variables to predict the presence or absence of individual fish species at individual sites. Second, environmental data were compiled for all stream sites in Wisconsin, relatively few of which had fish data. These environmental data were utilized with the fish models to predict the occurrence of each species at each stream site in the state under current climate conditions. Finally, the models were rerun under three scenarios of increasing summer air and water temperatures for each species at each stream site to predict changes in statewide fish distribution in response to a warming climate.

Fish data

Fish species models were developed with fish and environmental data collected from 393 sites located on 282 Wisconsin streams and rivers (Fig. 1). Sites were located throughout the state and encompassed a wide range and combination of environmental conditions, from intermittent headwaters to the largest rivers, from high-gradient rocky to low-gradient marshy channels, from cold water to cool water to warm water and from relatively undisturbed forested to heavily modified agricultural and urbanized catchments (Lyons et al., 2001, 2009; Lyons, 2006). Each site was sampled for fishes once between 1995 and 1999 during summer low flows by daytime electrofishing, the specific type of electrofishing and the length of stream sampled depending on the width and depth of the site. Fish sampling is most effective and fish populations are least mobile in Wisconsin streams during summer low flows. In wadeable streams <3 m wide, a single-pulsed DC backpack electrofisher was used for a length of 100 m. In wadeable streams >3 m wide, a tow-barge electrofisher with three hand-held anodes was used for a length of 400 m or 35 times the mean channel width, whichever was shorter. In non-wadeable rivers, a boat-mounted, fixed-anode boom electrofisher with a single netter was used for a length of 1620 m. In all sizes of stream, an attempt was made to capture all fishes observed. Models were developed, however, only for the 50 species that were collected from at least 20 sites (>5%; Table I). These 50 species represented all the common species found in small to medium streams in Wisconsin, and many of the common species found in larger rivers (Becker, 1983; Lyons et al., 2000). Those species that were captured but not modelled were either rare statewide or limited to the largest rivers.

image

Figure 1. Map of the state of Wisconsin showing major streams, rivers and lakes and the location of the 393 sites used to develop the fish models. The inset map shows the location of Wisconsin in the U.S.A.

Download figure to PowerPoint

Table I.  List of the 50 fish species for which models were developed, with their number of occurrences (N) in the model development data set (393 sites). Also given are their ecological classifications for water temperature (Temp) [W, warm water; T, cool water (= transitional); C, cold water (Lyons et al., 2009)], fishery management (Mgmt), [N, non-game; B, non-game, widely sold and used for bait; G, game (Wisconsin Department of Natural Resources)], tolerance to urbanization (Toler) [S, sensitive; M, moderate; T, tolerant (Lyons, 1992, 2006; Lyons et al., 1996, 2001; Wang et al., 1997, 2000, 2001, 2003)] and stream size requirements (Size) [H, headwater; M, mainstem; R, river; U, ubiquitous (Lyons, 1996, 2006; Lyons et al., 2000, 2001)]
Scientific nameCommon nameNEcological classifications
TempMgmtTolerSize
Family Cyprinidae      
 Campostoma anomalumCentral stoneroller49WNSM
 C. oligolepisLargescale stoneroller32WNSM
 Cyprinella spilopteraSpotfin shiner67WNMR
 Cyprinus carpioCommon carp137WNTR
 Hybognathus hankinsoniBrassy minnow28TNSH
 Luxilus cornutusCommon shiner107WNMM
 Margariscus margaritaPearl dace29TNMH
 Nocomis biguttatusHornyhead chub65WBSM
 Notemigonus crysoleucasGolden shiner43WBTM
 Notropis dorsalisBigmouth shiner28WNMH
 N. heterolepisBlacknose shiner26TNSM
 N. percobromusCarmine shiner28WNSM
 N. stramineusSand shiner50WNMR
 Phoxinus eosNorthern redbelly dace43TNSH
 P. erythrogasterSouthern redbelly dace23WNSH
 Pimephales notatusBluntnose minnow69WNTM
 P. promelasFathead minnow65WBTH
 Rhinichthys cataractaeLongnose dace49TNSM
 R. obtususWestern blacknose dace81TNTH
 Semotilus atromaculatusCreek chub133TNTH
Family Catostomidae      
 Catostomus commersoniiWhite sucker267TBTU
 Hypentelium nigricansNorthern hog sucker81TNSR
 Moxostoma erythrurumGolden redhorse63WNSR
 M. macrolepidotumShorthead redhorse103WNSR
Family Ictaluridae      
 Ameiurus melasBlack bullhead63WGTM
 A. natalisYellow bullhead44WGTM
 Ictalurus punctatusChannel catfish55WGMR
 Noturus flavusStonecat29WNSM
Family Esocidae      
 Esox luciusPike74TGMU
Family Umbridae      
 Umbra limiCentral mudminnow145TNTH
Family Salmonidae      
 Salmo truttaBrown trout69CGSH
 Salvelinus fontinalisBrook trout55CGSH
Family Gadidae      
 Lota lotaBurbot26TNSR
Family Gasterosteidae      
 Culaea inconstansBrook stickleback68TNTH
Family Cottidae      
 Cottus bairdiiMottled sculpin70CNSH
Family Centrarchidae      
 Ambloplites rupestrisRock bass94WGSR
 Lepomis cyanellusGreen sunfish89WGTH
 L. gibbosusPumpkinseed53WGMM
 L. macrochirusBluegill102WGMU
 Micropterus dolomieuSmallmouth bass151WGSR
 M. salmoidesLargemouth bass52WGMU
 Pomoxis nigromaculatusBlack crappie49WGMR
Family Percidae      
 Etheostoma caeruleumRainbow darter23WNSM
 E. flabellareFantail darter46WNSH
 E. nigrumJohnny darter111TNMU
 E. zonaleBanded darter26WNSR
 Perca flavescensYellow perch75TGMR
 Percina caprodesLogperch65WNSR
 P. maculataBlackside darter50WNSR
 Sander vitreusWalleye99TGSR

All fish models were tested and validated with either independent fish data or a data re-sampling procedure of the model development data set. Nineteen of the models were tested with data from a different data set of 1235 sites on 693 Wisconsin streams and rivers. The 1235 sites were also located throughout the state and encompassed a wide range of environmental conditions, and each was sampled between 1998 and 2004 during summer low flows by daytime electrofishing. These samples, however, focused on game species with only three of the larger non-game species [Cyprinus carpio L., Catostomus commersonii (Lacépède) and Hypentelium nigricans (LeSueur)] consistently targeted, collected and identified. So this larger data set could be used for validation of only those three non-game species and the 16 game species (Table I). Of the remaining 31 species, the 18 for which occurrence was relatively high (N≥ 50 sites) were validated by a sub-setting procedure in which the model was developed from a randomly chosen group of 294 sites (75%) from the 393 site data set and then tested with the remaining 99 sites (25%). For the other 13 species, where frequency of occurrence was relatively low (20–49 sites), the model was developed with all 393 sites and then tested with a randomized re-sampling 10-fold cross-validation procedure using the same data (Steinberg & Colla, 1997).

Environmental data

Using GIS, streams in Wisconsin were assigned a variety of environmental attributes at multiple spatial scales, described in more detail in Brenden et al. (2006), Steen et al. (2008) and Lyons et al. (2009). First, all streams in the state that were mapped at the 1:100 000 scale (U.S. National Hydrography data; http://nhd.usgs.gov/) were partitioned into discrete segments, each segment extending from a tributary confluence or lake or impoundment outlet (or the stream's source) downstream to the next tributary confluence or lake or impoundment inlet (or the stream's mouth). The 86 898 km of stream in the state was divided into 35 748 segments with a mean length of 2·43 km. Then, 27 environmental characteristics for each segment were determined and assigned, including various aspects of proximity to lakes and large rivers, geology and topography, climate and land cover and use (Table II). These 27 variables were selected based on previous studies that identified key environmental factors explaining fish distribution in Wisconsin and the Laurentian Great Lakes region (Lyons, 1996; Wiley et al., 1997; Zorn et al., 2002; McKenna, 2005; Hughes et al., 2006; Steen et al., 2008, 2010; Lyons et al., 2009; Diebel et al., 2010).

Table II.  Environmental variables used to model fish habitat in Wisconsin streams. The data sources, calculations, and mapping procedures for the variables are described more fully in Brenden et al. (2006), Steen et al. (2008) and Lyons et al. (2009)
VariableDescriptionSpatial scale
  1. c, channel; r, local riparian; w, local catchment; rt, upstream riparian network; wt, upstream catchment.

Proximity to lakes and large rivers  
 Lake upChannel distance from segment to first lake or impoundment upstream (m)c
 Lake downChannel distance from segment to first lake or impoundment downstream (m)c
 River downChannel distance from segment to large river (≥sixth order) downstream (m)c
Geology and topography  
 GradientSlope of the stream segment (m km−1)c
 SinuosityMeandering index: actual segment length–straight line length (dimensionless)c
 Val slopeAverage slope of land surrounding segment (m km−1)r, w, rt, wt
 Soil permAverage permeability of soil surrounding segment (mm h−1)r, w, rt, wt
 Fine textLand area with fine-texture surficial geology (%)wt
 Med textLand area with medium-texture surficial geology (%)wt
 Crse textLand area with coarse-texture surficial geology (%)wt
 SandstoneLand area with sandstone bedrock (%)wt
 ShaleLand area with shale bedrock (%)wt
 CarbonateLand area with carbonate bedrock (%)wt
 MetamorphLand area with metamorphic bedrock (%)wt
 IgneousLand area with igneous bedrock (%)wt
 DarcyIndex of potential rate of groundwater movement (dimensionless)r, w, rt, wt
 AreaCatchment area, the total land surface area draining to segment (km2)wt
Climate (1991–2003 values, interpolated from 156 weather stations)  
 Deg daysMean number of growing degree days >10° Cw, wt
 July airMaximum air temperature in July (° C)w, wt
 Mean airMean annual air temperature (° C)w, wt
 PrecipMean annual precipitation (mm)w, wt
Land cover and use (1992 values from satellite imagery)  
 AgricLand area that is cropland including orchards and hayfields (%)r, w, rt, wt
 UrbanLand area that is classified as residential, commercial or transportation (%)r, w, rt, wt
 GrassLand area that is open non-wooded, non-wetland, including pasture (%)r, w, rt, wt
 ForestLand area that is predominantly wooded and non-wetland (%)r, w, rt, wt
 WetlandLand area that is seasonally wet, including swamp (forested) and marsh (open) (%)r, w, rt, wt
 WaterOpen-water areas (lakes, ponds, impoundments and river channels) (%)r, w, rt, wt
Predicted stream flow  
 Aug low90% exceedence flow (i.e. low flow) for month of August (m3 s−1)c
 Aug yield90% exceedence flow for August per unit catchment area (m3 s−1 km−2)c
 Apr high10% exceedence flow (i.e. high flow) for month of April (m3 s−1)c
 Apr yield10% exceedence flow for April per unit catchment area (m3 s−1 km−2)c
 Year med50% exceedence flow (i.e. median flow) for entire year (m3 s−1)c
 Year yield50% exceedence flow for year per unit catchment area (m3 s−1 km−2)c
Predicted water temperature (1990–2002)  
 Sum WT13 year average of daily water temperature from 1 June to 31 August (° C)c
 July WT13 year average of daily water temperature for July (° C)c
 MDM WT13 year average of the maximum daily mean water temperature each year (° C)c
 Max JulyMaximum July water temperature for the 13 year period (° C)c
 Max MDMMaximum value of maximum daily mean water temperature for 13 year period (° C)c

Many of these environmental characteristics were expressed at more than one spatial scale, resulting in a total of 58 assigned environmental variables (Table II). The channel scale pertained to features of the stream segment itself, such as the local stream gradient, and to features of the overall stream channel network, such as the distance, via stream channels, from the stream segment to the nearest upstream lake or impoundment. The local riparian scale included characteristics, mainly related to land cover and use, within 60 m of the centre line of the channel on both sides of the stream segment, for a total width of 120 m. The local catchment scale encompassed attributes of the land that drained directly to the stream segment including local land cover and use, climate and valley slope, and excluding land draining to stream segments further upstream. The upstream riparian trace scale covered elements such as land cover and use, riparian zone slope and soil permeability in all riparian areas (i.e. within 60 m of either side of the channel) upstream of the segment, including tributaries. The upstream catchment scale described aspects of the entire catchment that drained to the stream segment, including land cover and use, surficial and bedrock geology, climate, soil permeability and valley slope.

Several of the 58 assigned environmental variables were used to develop statistical models to estimate 11 additional variables related to stream flow and water temperature for each segment for a total of 69 environmental variables used to model stream fish distribution (Table II). Water temperature and flow data are lacking for >99% of Wisconsin stream segments, including most of the model development sites, so it was impossible to use measured flows and temperatures. Stream flow variables were estimated from an unpublished model (L. Hinz, unpubl. data) based on a regression analysis of selected segment geology and topography, climate and land cover and use variables with daily flow data collected at 54 U.S. Geological Survey continuous flow gauging stations over a 20 year period (1981–2000) on a variety of streams and rivers throughout Wisconsin. The analysis predicted three exceedence flows, stream flows that were exceeded a certain percentage of days within a particular time period. These three exceedence flows are important predictors of species occurrence and abundance in the Laurentian Great Lakes region (Zorn et al., 2002; Steen et al., 2008, 2010). The August 90% exceedence flow was exceeded for 90% of the days in the month of August (27 days) in a typical year and was a measure of summer low flows. Conversely, the April 10% exceedence flow was exceeded for 10% of the days in the month of April (3 days) and was a measure of spring high flows. The annual 50% exceedence flow was exceeded for 50% of the days in the year (183 days) and was a measure of average stream flow. The three regression equations had r2 values of 90–97% and yielded predicted exceedence flows within 10% of true flows in ≥75% of independent test cases from 24 additional gauging stations. The three exceedence flows were expressed both in their direct form and per unit catchment area for a total of six flow variables.

Stream water temperatures were estimated from an artificial neural net model of measured water temperatures coupled with segment geology and topography, climate and land cover and use variables. This model was described in more detail in Roehl et al. (2006), Stewart et al. (2006) and Lyons et al. (2009). The model was based on single summers of water temperature data from 223 segments on streams throughout Wisconsin collected from 1990 to 2002 and matched with statewide air temperature data for the entire 13 year period from 156 weather stations. This time frame encompassed unusually hot (e.g. 1995), cold (1992), wet (1993) and dry (2002) years as well as more typical years and thus covered the range of current climatic conditions in the state. The model predicted daily mean water temperature for each summer day (1 June to 31 August) for 1990 to 2002 for a total of 1196 daily water temperatures for each segment. These daily predications were summarized to generate five water temperature variables shown to be important in explaining fish distribution patterns in streams of the Laurentian Great Lakes region (Wehrly et al., 2003; Steen et al., 2008; Lyons et al., 2009): average (over 13 years) June–August mean, average July mean, average maximum daily mean, maximum of the 13 July means and maximum of the 13 annual maximum daily means. In validation tests with independent water temperature data from additional segments, the model explained 67% of the variation in observed daily mean water temperatures for 31 segments, and predictions of July mean water temperatures for individual years from the model were within 2° C of observed temperatures for 55% and within 4° C for 80% of 171 segments. Prediction errors were unbiased, i.e. model results in any particular year were as likely to be too low as they were to be too high. Thus, over the 13 year water temperature modelling time frame, negative and positive temperature deviations would tend to cancel out, and predicted 13 year summaries of water temperatures would tend to be much closer to actual 13 year summaries than would be the case for predicted and actual temperatures in an individual year.

Fish species models

Fish species distribution models were based on classification-tree analysis applied via CART 5.0 software (Salford Systems, http://salford-systems.com/cart.php; Breiman et al., 1984; Steinberg & Colla, 1997). These models used the 69 environmental variables to predict the presence and absence of each of the 50 fish species. Classification-tree models are relatively simple to comprehend and interpret (Bell, 1999; De’ath & Fabricius, 2000) and have proven accurate and useful in a variety of fish distribution modelling contexts in North America (Olden & Jackson, 2002; Turgeon & Rodriguez, 2005; Steen et al., 2006, 2008; Brewer et al., 2007; Sowa et al., 2007; Hayer et al., 2008; Sharma & Jackson, 2008).

The output from a classification-tree model of species distribution is a series of trees with increasing numbers of branches. These branches are splits of the data into two groups, one in which the species is predicted to be present and the other in which it is predicted to be absent. A key issue in model development and selection is optimizing the size of the tree to maximize overall model accuracy and sensitivity (correct prediction of presence) and specificity (correct prediction of absence) while maintaining ecological realism and interpretability (Olden & Jackson, 2002). Models with too many branches are over-fit and may have high accuracy for the model development data set but poor performance in validation tests Over-fit models may also have high sensitivity but low specificity (or vice versa), and are more likely to contain spurious and misleading relations between species distribution and environmental variables (Bell, 1999).

In this study, the following process was used to determine optimum tree size and to select the final model. First, each species was modelled with two different statistical algorithms available in CART 5.0, Gini and Entropy, and both sets of trees were examined. A tree was considered good if overall accuracy, sensitivity and specificity were each >60% for both the model development and the validation data sets, fair if one or more of these measures was 50–60% and poor if one or more measures was <50%. Only good trees were considered further, unless all trees were no better than fair. Then the best tree for the Gini and the Entropy algorithms were compared. For 34 of 50 fish species models, one algorithm produced a clearly superior tree and that was chosen for the final model. In the 16 cases where the choice was less certain, two criteria were applied. First, relatively easy-to-interpret trees with four to 10 branches (Steen et al., 2008) and ecologically meaningful splits that reflected known habitat requirements and preferences of the species (Becker, 1983; Lyons et al., 2000) were selected. More or less complex trees or trees with counterintuitive or difficult to interpret splits were rejected, even if the latter had slightly better accuracy, sensitivity or specificity. Eleven species models were finalized based on this criterion. For the remaining five species models, the tree with the most similar values for sensitivity and specificity for the validation data set was chosen over the tree with more disparate values even if the tree with more disparate values had slightly better overall accuracy. This second criterion was applied because a primary goal was to be able to predict both species presence and species absence at new sites as accurately as possible rather than maximizing the prediction of presence at the expense of absence, or vice versa. So, for example, the finalized tree for fantail darter Etheostoma flabellare Rafinesque was based on the Entropy algorithm and had sensitivity of 64·6%, specificity of 70·6% and overall accuracy of 69·7%, whereas an alternate tree based on the Gini algorithm that was not chosen had sensitivity of 60·7%, specificity of 72·0% and overall accuracy of 70·6%. Once a final model had been chosen, the statistical significance of its overall accuracy was tested with Cohen's Kappa (Mantel et al., 2001). This test determined whether model predictions of the presence and absence of a fish species at the study sites were more accurate than random assignment of each individual site to either the presence or absence category based on the original proportions of sites in each category.

Predicting fish species distributions

Once a final species model had been selected, it was applied to each of the stream segments in the state using the appropriate environmental variables to determine whether or not the species was predicted to be present. The classification-tree models gave a probability of occurrence for each of the final branches, but for simplicity, occurrence was treated as a binary variable, with the greater of the two probabilities in a pair of branches indicating presence and the lesser indicating absence. Model predictions were then constrained in four ways, described below, to provide more realistic predictions. After these constraints were applied, the predicted distribution for each species was mapped and the total length of stream in which the species was predicted to occur was determined.

The constraints, covering location in the state, water temperature, stream size and land-use variables, represent a balancing between model ecological realism and statistical performance. These four classes of variables are well-documented as key determining factors for stream fish distribution, but they all did not appear in the models for all species for a variety of reasons. By treating these variables as constraints on model output, unrealistic predictions of occurrence were eliminated in an objective and justifiable fashion. Constraint thresholds were set at conservative levels so that only predictions that were clearly and unambiguously erroneous would be corrected.

The first model constraint dealt with the zoogeography of species. Because the models did not explicitly take into account variables related to the location of specific river systems or barriers to movement, they sometimes predicted species presence in drainage basins where the species did not occur (Diebel et al., 2010). In particular, many non-game species were absent from the Lake Superior basin because of natural barriers to colonization subsequent to the most recent glacial period even though apparently suitable habitat currently occurs there (Becker, 1983; Lyons et al., 2000). If a species that was known to be absent from a particular basin was predicted to occur in segments within that basin, then those segments were changed from present to absent.

The second constraint on model predictions related to water temperature tolerance. Although five measures of water temperature were included among the environmental variables, not all models included water temperature variables (Table III). This was because water temperature was correlated with several of the geological, topographic and climatic variables, and those surrogate variables alone sometimes produced a more accurate model than one that included any of the water temperature variables. In those cases, the surrogate variables presumably represented key thermal limits better than the five water temperature variables included in the analysis. Nonetheless it is well established that all fish species have upper and lower lethal water temperature thresholds, and because of the objectives of this study, it was essential that relevant water temperature effects were taken into account during the prediction of species distributions. Consequently, if a species was predicted to occur in segments where modelled water temperature was clearly unsuitable based on previous studies, then those segments were changed from present to absent. Segments in which average maximum daily mean water temperatures were warmer than 24·6° C were considered unsuitable for cold-water species (Table I), whereas segments colder than 20·7° C were considered unsuitable for warm-water species (Lyons et al., 2009). For cool-water (= transitional) species, segments with average maximum daily mean water temperatures colder than 18·8° C or warmer than 27·1° C were classified as unsuitable.

Table III.  Characteristics of the final fish species models. The number of terminal branches (also known as terminal nodes) in the final classification tree is given, and whether water temperature or air temperature variables or both were included in the final model is indicated
Fish speciesBranchesAir or water temperature in model?Environmental variables
1st split2nd split2nd split
  1. 1st split, the variable (Table II) used to make the first set of branches in the model; this is the single variable in the final model that best explains the distribution pattern of the species. The spatial scale of the variable is given after a hyphen; 2nd split, the variables (Table II) used to make the next two sets of branches in the model after the first split; these are next most important variables in the final model; NA, no second split was made and that the branch from the first split was a terminal branch. c, channel; Crse text, coarse-texture surficial geology; MDM, maximum daily mean; r, local riparian; rt, upstream riparian network; w, local catchment; wt, upstream catchment.

Family Cyprinidae     
 Campostoma anomalum4NoForest-wtSoil perm-wtNA
 C. oligolepis8WaterMax MDM-cDarcy-wtPrecip-wt
 Cyprinella spiloptera5AirYear med-cArea-wtMean air-w
 Cyprinus carpio5AirArea-wtMean air-wNA
 Hybognathus hankinsoni3NoWater-rForest-wtNA
 Luxilus cornutus6BothDeg days-wYear med-cCarbonate-wt
 Margariscus margarita7AirYear med-cAug yield-cVal slope-w
 Nocomis biguttatus6BothDeg days-wDarcy-wVal slope-r
 Notemigonus crysoleucas7NoLake down-cWetland-wtNA
 Notropis dorsalis7WaterYear med-cForest-wtNA
 N. heterolepis5AirMean air-wWater-rSoil perm-w
 N. percobromus5WaterMax July-cApr high-cNA
 N. stramineus5AirMean air-wArea-wtNA
 Phoxinus eos6AirDeg days-wArea-wtNA
 P. erythrogaster5WaterWetland-wtSoil perm-rGrass-wt
 Pimephales notatus9WaterForest-wtGrass-wSoil perm-rt
 P. promelas9AirArea-wtVal slope-wtForest-rt
 Rhinichthys cataractae9AirArea-wtMean air-wtNA
 R. obtusus8BothArea-wtDeg days-wtNA
 Semotilus atromaculatus5AirArea-wtJuly air-wNA
Family Catostomidae     
 Catostomus commersonii12BothApr high-cArea-wtNA
 Hypentelium nigricans5AirApr high-cMean air-wtNA
 Moxostoma erythrurum10AirYear med-cSoil perm-rVal slope-w
 M. macrolepidotum6AirArea-wtArea-wtNA
Family Ictaluridae     
 Ameiurus melas3WaterYear med-cSum WT-cNA
 A. natalis6AirForest-rtYear med-cVal slope-rt
 Ictalurus punctatus3AirArea-wtJuly air-wtNA
 Noturus flavus3WaterSum WT-cYear med-cNA
Family Esocidae     
 Esox lucius5NoArea-wtWater-wVal slope-w
Family Umbridae     
 Umbra limi7AirArea-wtWetland-wtWetland-rt
Family Salmonidae     
 Salmo trutta3NoApr high-cSandstone-wtNA
 Salvelinus fontinalis10AirYear med-cMean air-wtNA
Family Gadidae     
 Lota lota2AirMean air-wtNANA
Family Gasterosteidae     
 Culaea inconstans7AirArea-wtApr yield-cNA
Family Cottidae     
 Cottus bairdii7BothArea-wtMDM WT-cDeg days-w
Family Centrarchidae     
 Ambloplites rupestris4NoLake up-cYear med-cNA
 Lepomis cyanellus7AirMean air-wtSandstone-wtNA
 L. gibbosus11AirDeg days-wLake up-cVal slope-rt
 L. macrochirus3AirAug low-cDeg days-wtNA
 Micropterus dolomieu8BothArea-wtSum WT-cYear med-c
 M. salmoides4AirApr high-cMean air-wtNA
 Pomoxis nigromaculatus5AirArea-wtJuly air-wNA
Family Percidae     
 Etheostoma caeruleum4AirCrse text-wtForest-wtNA
 E. flabellare4NoYear med-cSoil perm-rNA
 E. nigrum7BothArea-wtUrban-wtJuly air-w
 E. zonale5BothMax MDM-cJuly air-wNA
 Perca flavescens6BothVal slope-rtMean air-wNA
 Percina caprodes11BothMean air-wtYear med-cSum WT-c
 P. maculata10BothMax MDM-cMean air-wNA
 Sander vitreus7AirArea-wtCarbonate-wtForest-w

The third constraint dealt with stream size preferences. Many Wisconsin species show clear stream size habitat requirements, some are only able to persist in small headwaters and tributaries and others only in large rivers (Lyons, 1996, 2006; Lyons et al., 2001). For the same reasons that water temperature variables were not always included, these stream size habitat requirements were also not reflected in some of the final models. All species in the analysis were classified by stream size preference based on previous studies (Table I; Becker, 1983; Lyons, 1996, 2006; Lyons et al., 2000, 2001). Headwater species were those found only in small streams, and they were considered absent from segments with a catchment area that exceeded 1500 km2 regardless of model predictions. River species were those occurring only in larger streams and rivers and were classified as absent if catchment area was <50 km2. Mainstem species were widely distributed, but were not found in the smallest streams or in the largest rivers. They were considered absent from segments with a catchment area <10 or >12 000 km2. Species classified as ubiquitous were not constrained by stream size.

The final constraint related to tolerance of human activities and disturbance in the catchment. Species varied in their ability to persist in the face of catchment land-use changes, particularly urbanization. Again, this was sometimes not reflected in the final model outputs. Based on previous studies of species tolerance (Lyons, 1992, 2006; Lyons et al., 1996, 2001) and responses to catchment urbanization (Wang et al., 1997, 2000, 2001, 2003), species were classified as sensitive, moderate or tolerant (Table I). Sensitive species could not tolerate extensive urban development and were considered absent if catchment urbanization exceeded 30%. Moderate species were less sensitive, but were coded as absent if catchment urbanization exceeded 50%. Tolerant species were able to persist in highly urbanized streams and had no constraints for catchment urbanization.

After species distributions had been predicted and mapped under current air and water temperature conditions, the models were rerun with warmer air and water temperatures. Three climate warming possibilities were explored, limited warming, moderate warming and major warming. These three possibilities were taken from recent efforts to downscale continental climate predictions of global circulation models (GCM) from the Intergovernmental Panel on Climate Change (IPCC, 2007). The original values for 150 km square grids of the earth's surface were statistically downscaled to smaller 10 km square grids in order to make more precise predictions of climate change for Wisconsin (WICCI, 2009). The downscaling effort considered 15 different GCM over the next 50 years, each with three greenhouse gas-emission projections, for 45 different climate predictions. The limited warming scenario used here represented the smallest air temperature increase of the 45 GCM emission predictions [GCM CSIRO-M3K-0 under optimistic projections (IPCC projection B1) of future greenhouse gas emissions], the moderate warming scenario represented the median air temperature increase [GCM GISS-AOM with ‘business as usual’ (A1) projections of future greenhouse gas emissions] and the major warming scenario represented the second highest air temperature increase [GCM MIROC3-HIRES with accelerating projections (A2) of future greenhouse gas emissions]. The highest increase was substantially greater than all others and was considered an outlier and was not used. Under the limited warming scenario, all three air temperature variables were projected to increase by 1° C, which translated to an increase of 170 for the growing degree-day variable, and the five water temperatures were assumed to increase by 0·8° C, based on the approximate increase of 0·8° C in maximum water temperature for each 1° C increase in air temperature observed for streams in the adjacent state of Minnesota (Pilgrim et al., 1998). The relation between air and water temperatures used here was an oversimplification, and in reality not all streams would respond the same to air temperature increases. Existing climate change projections, however, were not detailed enough to use in the water temperature model, and this model also could not directly take into account how changes in precipitation, land-use and groundwater contributions to streams might ameliorate or exacerbate effects of air temperature increases on water temperature. Under the moderate warming scenario the three air temperature variables increased by 3° C, which translated to an increase of 500 for the degree-day variable, and the five water temperature variables were assumed to increase by 2·4° C. Under the major warming scenario the three air temperature variables increased by 5° C, which translated to an increase of 900 for the degree-day variable, and the five water temperature variables were assumed to increase by 4° C.

For the three climate warming scenarios, predictions were again constrained by zoogeography, temperature tolerance, stream size requirements and sensitivity to catchment urbanization of the species. For the 16 game species and the four non-game species widely sold and used for fishing bait (Table I), however, the zoogeography constraint was not applied. These 20 species have been regularly moved among waters by anglers, despite regulations prohibiting this, and if a warmer climate made habitat conditions suitable in new areas, it would be plausible that these species would eventually become established there through angler introductions. After model constraints had been taken into account, the predicted distribution of each species was mapped and the total length of stream in which each species was predicted to occur was summarized for each of the climate change scenarios.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References

The 50 fish species models differed greatly in structure and composition (Table III). The number of terminal branches in the final models ranged from two to 12 with a mean of 6·9. Forty-three species models included a water temperature or an air temperature variable; 11 models had both air and water temperature variables, 25 had just air temperature and seven had just water temperature. Nineteen different environmental variables were used to make the first split or branching point in the tree for the 50 fish models. These were the most important variables in explaining the distribution patterns of species. Catchment area (15 species) and predicted annual median flow (seven species) were the variables most frequently used in the first split. Forty different environmental variables were used for the second splits, and these represented the next most important variables. Once again, catchment area (five species) and predicted annual median flow (six species) were the variables most frequently used to make the splits.

Fish models were moderately to highly accurate for both the model development and validation data sets (Table IV), and all final models explained significantly more of fish distribution patterns than would be expected by chance (Cohen's Kappa, P < 0·001). From the model development step, 48 models were classified as good and two as fair [Ambloplites rupestris (Rafinesque) and Micropterus salmoides (Lacépède)]. Overall accuracy ranged from 61·1 (M. salmoides) to 96·4% [Moxostoma macrolepidotum (Lacépède)] with a mean of 83·0%. Specificity varied from 53·2 (A. rupestris) to 95·9% (M. macrolepidotum) and averaged 80·1%. Sensitivity was particularly high, ranging from 73·0 [Ameiurus melas (Rafinesque)] to 100% (seven species) with a mean of 94·3%. For the validation step, 43 models were classified as good and seven as fair [Notemigonus crysoleucas (Mitchill), Notropis dorsalis (Agassiz), Pimephales promelas Rafinesque, Esox lucius L., Salmo trutta L., Lepomis gibbosus (L.) and Micropterus dolomieu Lacépède]. Overall accuracy ranged from 56·0 (N. crysoleucas) to 93·5% [Ictalurus punctatus (Rafinesque)] with a mean of 77·0%. Specificity varied from 53·3 (S. trutta) to 96·5% (M. dolomieu) and averaged 77·7%. Sensitivity ranged from 50·9 (L. gibbosus) to 90·9% (M. macrolepidotum) with a mean of 70·2%.

Table IV.  Fish species model overall per cent accuracy and per cent correct prediction of absence (specificity) and of presence (sensitivity) for the model development and validation data sets
Fish speciesDevelopmentValidation
Overall (%)Absence (%)Presence (%)Overall (%)Absence (%)Presence (%)
Family Cyprinidae      
 Campostoma anomalum76·674·193·973·573·871·4
 C. oligolepis84·082·510080·782·362·5
 Cyprinella spiloptera91·689·997·087·989·582·6
 Cyprinus carpio91·189·893·483·783·982·1
 Hybognathus hankinsoni73·372·189·375·573·475·0
 Luxilus cornutus83·681·087·975·876·774·4
 Margariscus margarita90·289·010082·884·370·0
 Nocomis biguttatus77·172·492·368·768·170·4
 Notemigonus crysoleucas74·372·390·756·056·651·2
 Notropis dorsalis77·475·610076·378·153·6
 N. heterolepis84·082·810081·281·773·1
 N. percobromus79·979·289·371·272·160·7
 N. stramineus82·481·986·076·878·764·0
 Phoxinus eos86·584·910083·284·374·4
 P. erythrogaster86·085·110086·387·665·2
 Pimephales notatus82·577·298·679·881·872·7
 P. promelas86·283·893·871·77658·3
 Rhinichthys cataractae90·589·893·977·881·863·6
 R. obtusus85·179·997·576·875·380·8
 Semotilus atromaculatus86·981·792·575·874·177·8
Family Catostomidae      
 Catostomus commersonii84·789·782·467·563·569·3
 Hypentelium nigricans75·369·298·879·380·171·1
 Moxostoma erythrurum90·988·798·475·878·863·2
 M. macrolepidotum96·495·997·189·989·490·9
Family Ictaluridae      
 Ameiurus melas75·175·573·074·474·970·2
 A. natalis67·397·782·573·482·580·0
 Ictalurus punctatus81·479·692·793·594·075·8
 Noturus flavus75·674·486·279·878·988·9
Family Esocidae      
 Esox lucius66·460·293·279·482·259·7
Family Umbridae      
 Umbra limi84·781·590·372·666·381·2
Family Salmonidae      
 Salmo trutta79·982·169·664·353·389·3
 Salvelinus fontinalis87·886·198·267·064·571·6
Family Gadidae      
 Lota lota72·270·696·273·569·888·5
Family Gasterosteidae      
 Culaea inconstans86·583·197·176·876·178·1
Family Cottidae      
 Cottus bairdii84·780·098·673·771·680·0
Family Centrarchidae      
 Ambloplites rupestris63·653·296·875·776·568·7
 Lepomis cyanellus84·582·691·071·973·466·0
 L. gibbosus77·173·898·178·280·950·9
 L. macrochirus65·568·866·466·060·465·0
 Micropterus dolomieu92·593·792·175·396·554·1
 M salmoides61·157·882·771·172·961·5
 Pomoxis nigromaculatus73·871·887·978·379·067·1
Family Percidae      
 Etheostoma caeruleum81·781·191·377·678·169·6
 E. flabellare76·075·578·369·770·664·3
 E. nigrum82·582·982·073·775·471·1
 E. zonale80·779·310078·479·069·3
 Perca flavescens74·369·893·375·777·460·5
 Percina caprodes88·787·692·377·877·080·0
 P. maculata92·492·094·074·777·068·0
 Sander vitreus83·399·091·292·768·380·6

Under current climate conditions, the statewide distribution of the 50 species varied greatly (Table V). Predicted occurrences ranged from a low of 3614 km, representing 4·2% of the total stream length in the state, for Notropis stramineus (Cope), to a high of 68 228 km, representing 78·5% of total stream length, for Semotilus atromaculatus (Mitchill). Seven species were expected to be found in >50% and 17 in <10% of the total stream length in the state. A variety of different predicted distribution patterns was evident, including species that were ubiquitous, e.g. C. commersonii [Fig. 2(a)], or limited to larger rivers, e.g. Sander vitreus (Mitchill) [Fig. 2(b)], the northern portion of the state, Phoxinus eos (Cope) [Fig. 2(c)], or the southern portion of the state, Phoxinus erythrogaster (Rafinesque) [Fig. 2(d)].

Table V.  Predictions from the 50 fish species models of the stream length and as a percentage of the total stream length in Wisconsin (86 898 km) that would be suitable for 50 fish species under current air and water temperatures, and predictions of the lengths of suitable stream and the per cent change from current climate conditions under three climate warming scenarios. Superscripts indicate the thermal classification of each species (1, warm water; 2, cool water; 3, cold water)
Fish speciesCurrent climateClimate warming scenarios
Limited warmingModerate warmingMajor warming
Length (km)Per cent of totalLength (km)Per cent changeLength (km)Per cent changeLength (km)Per cent change
Family Cyprinidae        
 Campostoma anomalum11018811·7101880101880101880
 C. oligolepis11168613·413966+16·118185+55·618210+55·8
 Cyprinella spiloptera148715·65898+21·16528+34·06528+34·0
 Cyprinus carpio181839·49534+16·510577+29·310577+29·3
 Hybognathus hankinsoni24449351·244295−0·442543−4·431117−30·1
 Luxilus cornutus12715631·319294−29·06244−77·06236−77·0
 Margariscus margarita22329826·816479−28·14556−80·415−99·9
 Nocomis biguttatus12080423·917055−18·09145−56·09365−55·0
 Notemigonus crysoleucas11480417·0148040148040148040
 Notropis dorsalis1927810·711145+20·111324+22·111324+22·1
 N. heterolepis21723819·810729−37·8778−95·5331−98·1
 N. percobromus179449·19743+22·610868+36·810868+36·8
 N. stramineus136144·25496+52·18180+126·38180+126·3
 Phoxinus eos23512240·419798−43·687−99·82−100
 P. erythrogaster11404016·221923+56·121923+56·121923+56·1
 Pimephales notatus11627718·716695+2·616695+2·616695+2·6
 P. promelas14173448·039935−4·334374−17·634374−17·6
 Rhinichthys cataractae21526317·611269−26·23022−80·2902−94·1
 R. obtusus25873067·636554−37·815084−74·311973−79·6
 Semotilus atromaculatus26822878·559664−12·648617−28·736270−46·8
Family Catostomidae        
 Catostomus commersonii22915933·623246−20·311409−60·95085−82·6
 Hypentelium nigricans21127913·09871−12·56976−38·22269−79·9
 Moxostoma erythrurum145255·26229+37·712916+185·415642+245·6
 M. macrolepidotum166247·67103+7·27103+7·27103+7·2
Family Ictaluridae        
 Ameiurus melas11807020·824536+35·833286+84·233300+84·3
 A. natalis1916510·518699+104·022694+147·623319+154·4
 Ictalurus punctatus148615·65680+16·86446+32·66446+32·6
 Noturus flavus139404·516643+322·426415+570·426959+584·2
Family Esocidae        
 Esox lucius21527517·614379−5·910114−33·84229−72·3
Family Umbridae        
 Umbra limi25490163·251128−6·843640−20·522553−58·9
Family Salmonidae        
 Salmo trutta33724142·934296−7·924908−33·14378−88·2
 Salvelinus fontinalis32880233·116245−43·61618−94·40−100
Family Gadidae        
 Lota lota274478·63478−53·30−1000−100
Family Gasterosteidae        
 Culaea inconstans26099870·259392−2·652844−13·436898−39·3
Family Cottidae        
 Cottus bairdii35959968·646547−21·920936−64·92755−95·4
Family Centrarchidae        
 Ambloplites rupestris11375915·8137590137590137590
 Lepomis cyanellus12208025·424808+12·436368+64·751526+133·4
 L. gibbosus11309415·113657+4·315121+15·515046+14·9
 L. macrochirus11759220·217746+0·917746+0·917746+0·9
 Micropterus dolomieu181729·48172010905+33·416094+96·9
 M salmoides12013423·227031+34·327031+34·327031+34·3
 Pomoxis nigromaculatus1934210·811877+27·113984+49·713984+49·7
Family Percidae        
 Etheostoma caeruleum141114·75638+37·15638+37·15638+37·1
 E. flabellare13234237·2323420323420323420
 E. nigrum26032669·458827−2·548699−19·339277−34·9
 E. zonale144465·16435+44·77842+76·47842+76·4
 Perca flavescens286449·98117−6·12045−76·30−100
 Percina caprodes186129·98612010374+20·513217+53·5
 P. maculata178639·06777−13·85755−26·85755−26·8
 Sander vitreus248325·64526−6·32004−55·7600−87·6
image

Figure 2. Predicted distribution of four representative fishes under current climate conditions. Only stream segments where the species is predicted to occur are shown. (a) Catostomus commersonii, a ubiquitous species, (b) Sander vitreus, a species limited to larger rivers, (c) Phoxinus eos, a species found primarily in northern Wisconsin and (d) Phoxinus erythrogaster, a species found primarily in southern Wisconsin.

Download figure to PowerPoint

Predicted responses to climate warming varied dramatically among the 50 species. Overall, 23 species declined in distribution, 23 increased and four had no change (Table V), but the losses in habitat by declining species were much greater than the gains in habitat by increasing species. The predicted total combined loss of stream length inhabited by the declining species was 126 441 km (5497 km per species) for the limited warming scenario, 343 034 km (14 915 km per species) for the moderate warming scenario and 484 048 km (21 046 km per species) for the major warming scenario. The total combined gain in stream length inhabited by the increasing species was 66 605 km (2896 km per species) for the limited warming scenario, 127 488 km (5543 km per species) for the moderate warming scenario and 154 537 km (6719 km per species) for the major warming scenario.

Not surprisingly, responses to climate warming differed among the three thermal guilds of fishes. All three of the cold-water species [S. trutta, Salvelinus fontinalis (Mitchill) and Cottus bairdii Girard] declined substantially with increasing air and water temperature, with S. fontinalis extirpated from Wisconsin streams under the major warming climate change scenario. Because S. trutta lacked a water or air temperature component in its model, all its predicted decline was due to the water temperature constraint. Declines in the other two cold-water species were primarily caused by limiting factors in their models rather than the constraints. All 16 of the cool-water species were also predicted to decrease in distribution as climate warmed, with Lota lota (L.) gone from Wisconsin streams under both the moderate and major warming scenarios and P. eos and Perca flavescens (Mitchill) gone under the major warming scenario. Because Hybognathus hankinsoni Hubbs and Esox lucius L. lacked a water or air temperature component in their models, all their predicted declines were due to the water temperature constraint, but declines in the other 14 cool-water species were mainly due to factors within their particular species model. Among the 31 warm-water species, four species were predicted to decline with warmer conditions [Luxilus cornutus (Mitchill), Nocomis biguttatus (Kirtland), P. promelas and Percina caprodes (Rafinesque)], four species showed no change in distribution [Campostoma anomalum (Rafinesque), N. crysoleucas, A. rupestris and E. flabellare] and the remaining 23 species increased in distribution. Of the 23 increasing species, predicted gain varied dramatically from 154 km (0·9% gain) for Lepomis macrochirus Rafinesque, 418 km (2·6%) for Pimephales notatus (Rafinesque) and 479 km (7·2%) for M. macrolepidotum to 23 019 km (584·2%) for Noturus flavus Rafinesque and 29 446 km (133·4%) for Lepomis cyanellus Rafinesque under the major warming scenario. The zoogeography constraint limited the predicted increases of 10 species whereas the stream size constraint limited the predicted increase of seven species. The land-use constraint did not substantially limit increases of any species at a statewide scale, but did restrict predicted local increases of eight species near major urban areas.

Spatial patterns of distribution change in response to warming temperatures also varied among fishes. For species that declined in distribution, the tendency was for the predicted range of species to remain roughly stable until the species was nearly extirpated, but for the density of segments predicted to contain the species to decrease steadily with increasingly warm temperatures, as illustrated by the cold-water C. bairdii (Fig. 3) and the cool-water H. nigricans (Fig. 4). For the warm-water species that increased in distribution, two primary patterns were evident. First, some species expanded northwards as climate warmed, especially if they were gamefishes without zoogeographic constraints on potential distribution, as illustrated by Pomoxis nigromaculatus (LeSeur) (Fig. 5). Second, the density of segments predicted to contain the species increased as the species expanded into many more streams within areas where they occurred already, as illustrated by N. flavus (Fig. 6). The new streams that were occupied were generally relatively small and prior to climate warming had been too cold for the species.

image

Figure 3. Predicted distribution of Cottus bairdii, a cold-water species, under four climate warming scenarios: (a) current conditions, (b) limited warming, (c) moderate warming and (d) major warming. Only stream segments where the species is predicted to occur are shown.

Download figure to PowerPoint

image

Figure 4. Predicted distribution of Hypentelium nigricans, a cool-water species, under four climate warming scenarios: (a) current conditions, (b) limited warming, (c) moderate warming and (d) major warming. Only stream segments where the species is predicted to occur are shown.

Download figure to PowerPoint

image

Figure 5. Predicted distribution of Pomoxis nigromaculatus, a warm-water species, under four climate warming scenarios: (a) current conditions, (b) limited warming (c) moderate warming and (d) major warming. Only stream segments where the species is predicted to occur are shown.

Download figure to PowerPoint

image

Figure 6. Predicted distribution of Noturus flavus, a warm-water species, under four climate warming scenarios: (a) current conditions, (b) limited warming, (c) moderate warning and (d) major warming. Only stream segments where the species is predicted to occur are shown.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References

The findings from this study indicate that catchment-scale environmental variables organized within a GIS framework can be used to develop statistical models that accurately predict the distribution of fish species within Wisconsin streams. The most useful environmental variables in the Wisconsin models are related to stream size (catchment area and median annual stream flow), climate (summer air temperatures) and summer water temperatures. Overall accuracy, specificity and sensitivity of the Wisconsin models are similar to or better than values of comparable fish models developed using a variety of statistical approaches (e.g. artificial neural nets, logistic regression and linear discriminant analysis) for other temperate areas of North America, Europe, Australia and New Zealand (Aguilar-Ibarra et al., 2003; Joy & Death, 2004; McKenna, 2005; Drake & Lodge, 2006; Fransen et al., 2006; McKenna et al, 2006; Mugodo et al., 2006; Buisson et al., 2007; Steen et al., 2008, 2010; Diebel et al., 2010). Most of these other studies also found stream size, climate and water temperature to be important predictor variables. Catchment-scale GIS-based models have been found to be useful in predicting effects of human activity, particularly land-use and climate changes, on the occurrence of fishes in streams and rivers throughout North America (Peterson & Kwak, 1999; Van Sickle et al., 2004; Flebbe et al., 2006; Kilgour & Stanfield, 2006; Chu et al., 2008; Nelson et al., 2009; Steen et al., 2010). The models developed in this study provide unique and valuable insights into the changes in fish distribution in Wisconsin streams expected in response to the air and water temperature increases likely to occur as a consequence of global climate change.

The Wisconsin models predict major declines in the occurrence of many fish species in response to a warmer climate. This response was expected for the three cold-water species, which are, by definition, limited by warmer water temperatures. Many previous studies have predicted substantial declines in the distribution and abundance of cold-water species such as salmonids and cottids with rising air and water temperatures (Meisner, 1990a,b; Keleher & Rahel, 1996; Lehtonen, 1996; Nakano et al., 1996; Daufresne et al., 2003; Flebbe et al., 2006; Battin et al., 2007; Williams et al., 2009; Steen et al., 2010). The Wisconsin models also predict that all 16 cool-water species would have reduced distributions with climate warming, and perhaps surprisingly, that many cool-water species would decline to the same or a greater extent than the cold-water species. Relatively few previous studies have examined cool-water species responses to climate change, and results from these studies have been variable. Most studies predicted that cool-water species would decline less than cold-water species (Magnuson et al., 1997; Mohseni et al., 2003) and that a few cool-water species might actually increase in abundance or distribution in certain areas (Lehtonen, 1996; Lappalainen & Lehtonen, 1997; Fang et al., 2004; Chu et al., 2005; Winfield et al., 2008; Steen et al., 2010). An early study of U.S. streams, however, predicted similar declines in occurrence for cold-water and cool-water fishes (Eaton & Scheller, 1996). Cool-water species are known to be less sensitive to increases in maximum water temperatures than cold-water species (Lyons et al., 2009), but the findings from the Wisconsin models suggest that some cool-water species may be similar to cold-water species in sensitivity to changes in other aspects of the thermal regime such as the mean summer water temperature or the duration of the growing season. Thus, in Wisconsin streams, cold-water and cool-water fishes may be similarly vulnerable to climate warming. Many of the cool-water species in this study are at or near the southern edge of their range in Wisconsin, and whether they would decline as greatly in areas further to the north is an important question for further research.

The Wisconsin models gave highly variable responses to climate warming for the 31 warm-water species, with four species predicted to decline in distribution, four to have no change and 23 to increase. Among the 23 increasing species, gains ranged from minimal to very large. Declines in the distribution of warm-water species in response to climate warming are somewhat counterintuitive, and may be a spurious result. The climate warming scenarios represent combinations of climate, water temperature and other environmental variables that did not exist within the data set used to build the models, and extrapolating beyond the model development conditions may yield unreliable results. This is a potential problem for all climate change modelling, as the expected future climate conditions have no current precedents, and there are many irresolvable uncertainties in terms of ecosystem response to climate change (Schindler et al., 2008; Johnson & Weaver, 2009). Thus, all projections and conclusions from this and previous studies should be viewed with caution. That being said, it is possible that some warm-water species will decline as climate warms, as has been proposed in several previous studies (Eaton & Scheller, 1996; Magnuson et al., 1997; Peterson & Kwak, 1999; Whitledge et al., 2006). In certain stream segments, water temperatures may become too warm for those warm-water species that prefer relatively cool-water temperatures compared to other warm-water species. The cold-water:cool-water:warm-water species classification is a simplification of a thermal continuum, and there are differences among species within each thermal guild in temperature tolerances (Eaton & Scheller, 1996; Lyons et al., 2009). The four Wisconsin warm-water species predicted to decline with climate warming are arrayed on the colder end of the thermal tolerance continuum within the warm-water species guild, and thus they may decrease while other warm-water species increase.

The four warm-water species predicted to have no change in their distributions in response to climate warming all lack water temperature or air temperature variables in their models. Thus, their predicted stable distribution may be unrealistic and caused by a limitation in model sensitivity to climate change. The absence of temperature variables in the model, however, may also reflect the relatively low importance of water and air temperatures as limiting factors for these species in Wisconsin streams under current conditions. It may be that other factors such as habitat volume, stream size and flow, as reflected in stream velocity and habitat structure (e.g. formation of riffle habitats), including stream gradient and valley slope variables, or channel morphology and bottom substratum, as reflected in geology and land cover variables, are more important than temperature in determining the distribution of these species in Wisconsin streams. It is likely, however, that water temperature is a more apparent limiting factor elsewhere in their range, and in those areas these species may experience a substantial change in distribution in response to climate warming.

In the same vein, for the 23 warm-water species predicted to become more widespread, the great range in the magnitude of increase among species can be explained by the relative importance of water temperature as a factor limiting each species distribution. Wisconsin streams that are currently too cold for warm-water fishes tend to be relatively small headwaters (Lyons, 1996; Lyons et al., 1996, 2009). Some warm-water species, such as N. flavus and L. cyanellus, are capable of living in small streams, but other warm-water species are not (see Table I). As these headwaters become warmer, warm-water species adapted for small-stream life are predicted to occupy them, leading to a relatively large increase in their distributions. These headwaters, however, remain unsuitable for warm-water species requiring relatively large stream and river habitats, such as P. nigromaculatus and M. macrolepidotum, because they are precluded by stream size rather than water temperature. As climate warms, these warm-water river species are predicted to expand into those relatively few river habitats that are currently cold water or cool water, but not into smaller streams, so their overall increase in distribution is modest.

For most non-game warm-water species, movement barriers further limit predicted increases in distribution. Many areas of northern Wisconsin are currently unsuitable for warm-water non-game species restricted to southern Wisconsin, such as P. erythrogaster. As climate warms, the models for many of the southern species predict that stream conditions will become suitable in northern Wisconsin, but the species will be unable to colonize these streams because of waterfalls and dams that block upstream movement towards the north. This is illustrated by N. flavus (Fig. 6). Presently, the model for this species predicts that habitat is unsuitable in a large portion of northern Wisconsin. As air and water temperatures increase, many streams in this area are predicted to become suitable, but the area is inaccessible to colonization because of a series of impassable barriers to upstream movement. This non-game species is not routinely used for bait in Wisconsin and is unlikely to be moved around barriers by anglers or government river management agencies. Thus, although N. flavus is expected to greatly expand its occurrence in southern and central Wisconsin as climate warms, it is predicted to remain absent from much of northern Wisconsin.

The models developed here are powerful new tools to explore potential impacts of climate change and other anthropogenic factors such as land-use changes on stream fishes, but they must be used with caution. Their greatest value lies in the level of detail and spatial precision they provide for predictions. Their strengths, however, should not blind readers to their limitations. They are ecological models, and as such represent major oversimplifications of the complex suite of variables that determine the suitability of a particular stream segment for a particular fish species. Although they have a relatively high level of accuracy compared to other similar models, they still misclassify from 6·5 to 44% of species occurrences under current conditions. Their use in predicting future fish distributions assumes that the observed relations between species and environmental variables are causative, which is probably not always to be the case. This modelling exercise also presumes that relations between species and their environment will remain constant with rising temperatures, which is unlikely (Davis et al., 1998). Finally, the models do not directly address biotic interactions (e.g. predation, competition and disease) that may limit species distributions, nor how these biotic interactions may change with a warming climate. Given these limitations, model predictions should be viewed as indices of the relative magnitude and spatial pattern of species distribution changes in response to a warming climate rather than absolute estimates of the exact amounts and locations of changes.

Along with the limitations of the species models the simplicity of the climate change scenarios that were explored must also be considered. Although the summer air temperature increases used here represent reasonable approximations of the range of warming predicted by a variety of climate models, it is certain that not all streams will experience the same increase in water temperature for a given air temperature increase as was assumed for the species models. For any increase in summer air temperature, streams will vary in their increase in water temperature based on their stream flow, relative groundwater inputs, stream channel morphology, solar radiation inputs and riparian and catchment vegetation and land use (Wehrly et al., 2009). Moreover, climate change will involve more than just increases in air temperature and will probably include changes in the amount, timing and form (i.e. rain v. snow) of precipitation and the intensity and duration of storm events and climate extremes (e.g. hot spells and droughts) (Magnuson et al., 1997). These changes will in turn influence stream flows, groundwater inputs, channel morphology, solar radiation, vegetation and land use and ultimately water temperatures and habitat suitability for fishes. Finally, climate warming will probably affect more than just summer water temperatures, including reducing the severity of winter water temperatures, which could improve fish overwinter survival, and altering the timing, rate and variability or spring water temperature rises and autumn water temperature declines, which could modify fish spawning, egg development and hatching and larval growth and survival. These other water temperature changes may be as or even more important than increases in summer water temperatures in determining stream habitat suitability and distribution patterns for some species. A new research initiative in Wisconsin is attempting to develop improved stream flow and water temperature models that are more sensitive to these other aspects of climate change. These new flow and temperature models will be applied to more detailed and realistic climate change scenarios to provide more accurate inputs for the species models in order to generate better predictions of climate change effects on fish species distributions.

Overall, the results of the current study predict that even small increases in summer air and water temperatures owing to climate warming will have major effects on the distribution of common stream fishes in Wisconsin. An equal number of species will increase as will decrease, but there will be a net loss of species habitat because other limiting environmental factors and barriers to dispersal and colonization will prevent expanding warm-water species from fully replacing declining cold-water and cool-water species. Consequently, some streams will experience a decline in species richness, as cold-water and cool-water fishes are lost but are not fully replaced by warm-water fishes. Fisheries opportunities will probably also decrease as S. trutta and S. fontinalis, popular cold-water gamefishes, disappear from small streams but warm-water gamefishes, none of which thrive in small streams, do not expand to take their place. Species changes will be most dramatic in small geographically isolated headwater streams in northern Wisconsin that currently have cold to cool summer water temperatures and are dominated by cold-water and cool-water fishes. There, cold-water and cool-water species will decline and eventually disappear as climate warms. Only a small subset of the warm-water species, however, will be able to replace them. These small streams lack suitable habitat for many warm-water species, and distance and barriers to movement will impede their colonization by southern Wisconsin warm-water species (e.g. P. erythrogaster, N. flavus and E. flabellare) for which the habitat would be suitable. Species changes are likely to be least in larger and warmer streams and rivers in southern Wisconsin that are already dominated by warm-water fishes.

The authors thank T. Brenden, A. Cooper, L. Hinz, P. Kanehl, A. Martin, C. Smith and S. Westenbroek for assistance in collecting, compiling and processing the data used to generate the species models and stream segment characteristics. Helpful reviews of earlier versions of this manuscript were provided by J. McKenna, J. Schaeffer and three anonymous reviewers. Support for this study was provided by a U.S. Environmental Protection Agency ‘Science to Achieve Results' Grant (R830596) through the National Center for Environmental Research; the U.S. Geological Survey, National GAP Analysis Program, Great Lakes Aquatic Gap Project; Federal Aid in Sport Fish Restoration Project F-80-R, Project F-95-P, studies SSMP and SSCN and the Wisconsin Department of Natural Resources.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References
  • Aguilar-Ibarra, A., Gevrey, M., Park, Y.-S., Lim, P. & Lek, S. (2003). Modeling the factors that influence fish guilds composition using a back-propagation network: assessment of metrics for indices of biotic integrity. Ecological Modelling 160, 281290.
  • Battin, J., Wiley, M. W., Ruckelhaus, M. H., Palmer, R. N., Korb, E., Bartz, K. K. & Imaki, H. (2007). Projected impacts of climate change on salmon habitat restoration. Proceedings of the National Academy of Sciences of the United States of America 104, 67206725.
  • Becker, G. C. (1983). Fishes of Wisconsin. Madison, WI: University of Wisconsin Press.
  • Bell, J. F. (1999). Tree-based methods. In Machine Learning Methods for Ecological Applications (Fielding, A. H., ed.), pp. 89105. Boston, MA: Kluwer.
  • Breiman, L., Friedman, J., Olshen, R. & Stone, C. (1984). Classification and Regression Trees. Pacific Grove, CA: Wadsworth.
  • Brenden, T. O., Clark, R. D., Cooper, A. R., Seelbach, P. W., Wang, L., Aichele, S. S., Bissell, E. G. & Stewart, J. S. (2006). A GIS framework for collecting, managing, and analyzing multiscape landscape variables across large regions for river conservations and management. In Landscape Influences on Stream Habitats and Biological Assemblages (Hughes, R. M., Wang, L. & Seelbach, P. W., eds), pp. 4974. Bethesda, MD: American Fisheries Society.
  • Brewer, S. K., Rabeni, C. F., Sowa, S. P. & Annis, G. (2007). Natural landscape and stream segment attributes influencing the distribution and relative abundance of riverine smallmouth bass in Missouri. North American Journal of Fisheries Management 27, 326341.
  • Buisson, L., Blanc, L. & Grenouillet, G. (2007). Modelling stream fish species distribution in a river network: the relative effects of temperature versus physical factors. Ecology of Freshwater Fish 17, 244257.
  • Chu, C., Mandrak, N. E. & Minns, C. K. (2005). Potential impacts of climate change on the distributions of several common and rare freshwater fishes in Canada. Diversity and Distributions 11, 299310.
  • Chu, C., Jones, N. E., Mandrak, N. E., Piggott, A. R. & Minns, C. K. (2008). The influence of air temperature, groundwater discharge, and climate change on the thermal diversity of stream fishes in southern Ontario watersheds. Canadian Journal of Fisheries and Aquatic Sciences 65, 297308.
  • Daufresne, M, Roger, M. C., Capra, H. & Lamouroux, N. (2003). Long-term changes within the invertebrate and fish communities of the Upper Rhône River: effects of climatic factors. Global Change Biology 10, 124140.
  • Davis, A. J., Jenkinson, L. S., Lawton, J. H., Shorrocks, B. & Wood, S. (1998). Making mistakes when predicting shifts in species range in response to global warming. Nature 391, 783786.
  • De’ath, G. & Fabricius, K. E. (2000). Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81, 31783192.
  • Diebel, M. W., Maxted, J. T., Jensen, O. P. & Vander Zanden, M. J. (2010). A spatial autocorrelative model for targeting stream restoration to benefit sensitive nongame fishes. Canadian Journal of Fisheries and Aquatic Sciences 67, 165176.
  • Drake, J. M. & Lodge, D. M. (2006). Forecasting potential distributions of nonindigenous species with a genetic algorithm. Fisheries 31, 916.
  • Eaton, J. G. & Scheller, R. M. (1996). Effects of climate warming on fish thermal habitat in streams of the United States. Limnology and Oceanography 41, 11091115.
  • Erickson, T. R. & Stefan, H. G. (2000). Linear air/water temperature correlation for streams during open water periods. Journal of Hydrologic Engineering 5, 317321.
  • Fang, X., Stefan, H. G., Eaton, J. G., McCormick, J. H. & Alam, S. R. (2004). Simulation of thermal/dissolved oxygen habitat for fishes in lakes under different climate scenarios. Part 1. Cool-water fish in the contiguous US. Ecological Modelling 172, 1337.
  • Ficke, A. D., Myrick, C. A. & Hansen, L. J. (2007). Potential impacts of global climate change on freshwater fisheries. Reviews in Fish Biology and Fisheries 17, 581613.
  • Fisher, W. L. & Rahel, F. J. (Eds) (2004). Geographic Information Systems in Fisheries. Bethesda, MD: American Fisheries Society.
  • Flebbe, P. A., Roghair, L. D. & Bruggink, J. L. (2006). Spatial modeling to project southern Appalachian trout distribution in a warmer climate. Transactions of the American Fisheries Society 135, 13711382.
  • Fransen, B. R., Duke, S. D., McWethy, L. G., Walter, J. K. & Bilby, R. E. (2006). A logistic regression model for predicting the upstream extent of fish occurrence based on geographical information systems data. North American Journal of Fisheries Management 26, 960975.
  • Hayer, C. A., Wall, S. S. & Berry, C. R. (2008). Evaluation of predicted fish distribution models for rare fish species in South Dakota. North American Journal of Fisheries Management 28, 12591269.
  • Hughes, R. M., Wang, L. & Seelbach, P. W. (Eds) (2006). Landscape Influences on Stream Habitats and Biological Assemblages. Bethesda, MD: American Fisheries Society.
  • Jackson, D. A. & Mandrak, N. E. (2002). Changing fish biodiversity: predicting the loss of cyprinid biodiversity due to global climate change. In Fisheries in a Changing Climate (McGinn, N. A., ed.), pp. 8998. Bethesda, MD: American Fisheries Society.
  • Jager, H. I., Van Winkle, W. & Holcomb, B. D. (1999). Would hydrologic climate changes in Sierra Nevada streams influence trout persistence? Transactions of the American Fisheries Society 128, 222240.
  • Johnson, T. B. & Evans, D. O. (1990). Size-dependent winter mortality of young-of-the-year white perch: climate warming and invasion of the Laurentian Great Lakes. Transactions of the American Fisheries Society 119, 301313.
  • Johnson, T. E. & Weaver, C. P. (2009). A framework for assessing climate change impacts on water and watershed systems. Environmental Management 43, 118134.
  • Joy, M. K. & Death, R. G. (2004). Predictive modelling and spatial mapping of freshwater fish and decapod assemblages using GIS and neural networks. Freshwater Biology 49, 10361052.
  • Keleher, C. J. & Rahel, F. J. (1996). Thermal limits to salmonid distributions in the Rocky Mountain Region and potential habitat loss due to global warming: a geographic information system (GIS) approach. Transactions of the American Fisheries Society 125, 113.
  • Kilgour, B. W. & Stanfield, L. W. (2006). Hindcasting references conditions in streams. In Influence of Landscapes on Stream Habitats and Biological Assemblages (Hughes, R. M., Wang, L. & Seelbach, P. W., eds), pp. 623640. Bethesda, MD: American Fisheries Society.
  • Lappalainen, J. & Lehtonen, H. (1997). Temperature habitats for freshwater fishes in a warming climate. Boreal Environment Research 2, 6984.
  • Lehtonen, H. (1996). Potential effects of global warming on northern European freshwater fish and fisheries. Fisheries Management and Ecology 3, 5971.
  • Lyons, J. (1996). Patterns in the species composition of fish assemblages among Wisconsin streams. Environmental Biology of Fishes 45, 329341.
  • Lyons, J. (2006). A fish-based index of biotic integrity to assess intermittent headwater streams in Wisconsin, USA. Environmental Monitoring and Assessment 122, 239258.
  • Lyons, J., Wang, L. & Simonson, T. D. (1996). Development and validation of an index of biotic integrity for coldwater streams in Wisconsin. North American Journal of Fisheries Management 16, 241256.
  • Lyons, J., Cochran, P. A. & Fago, D. (2000). Wisconsin Fishes 2000. Status and Distribution. Madison, WI: University of Wisconsin Sea Grant Institute.
  • Lyons, J., Piette, R. R. & Niermeyer, K. W. (2001). Development, validation, and application of a fish-based index of biotic integrity for Wisconsin's large warmwater rivers. Transactions of the American Fisheries Society 130, 10771094.
  • Lyons, J., Zorn, T., Stewart, J., Seelbach, P. W., Wehrly, K. & Wang, L. (2009). Defining and characterizing coolwater streams and their fish assemblages in Michigan and Wisconsin, USA. North American Journal of Fisheries Management 29, 11301151.
  • Magnuson, J. J., Crowder, L. B. & Medvick, P. A. (1979). Temperature as an ecological resource. American Zoologist 19, 331343.
  • Magnuson, J. J., Webster, K. E., Assel, R. A., Bowser, C. J., Dillon, P. J., Eaton, J. G., Evans, H. E., Fee, E. J., Hall, R. I., Mortsch, L. R., Schindler, D. W. & Quinn, F. H. (1997). Potential effects of climate changes on aquatic systems: Laurentian Great Lakes and Precambrian Shield Region. Hydrological Processes 11, 825871.
  • Mantel, S., Williams, H. C. & Ormerod, S. J. (2001). Evaluating presence-absence models in ecology; the need to account for prevalence. Journal of Applied Ecology 38, 921931.
  • Marshall, D. W., Fayram, A. H., Panuska, J. C., Baumann, J. & Hennessy, J. (2008). Positive effects of agricultural land use changes on coldwater fish communities in southwest Wisconsin. North American Journal of Fisheries Management 28, 944953.
  • McGinn, N. A. (Ed.) (2002). Fisheries in a Changing Climate. Bethesda, MD: American Fisheries Society.
  • McKenna, J. E. (2005). Application of neural networks to prediction of fish diversity and salmonid production in the Lake Ontario basin. Transactions of the American Fisheries Society 134, 2843.
  • McKenna, J. E., McDonald, R. P., Castiglione, C., Morrison, S. S., Kowalski, K. P. & Passino-Reader, D. P. (2006). A broadscale fish-habitat model development process: Genesee Basin, New York. In Landscape Influences on Stream Habitats and Biological Assemblages (Hughes, R. M., Wang, L. & Seelbach, P. W., eds). pp. 533544. Bethesda, MD: American Fisheries Society.
  • Meisner, J. D. (1990a). Effect of climatic warming on the southern margins of the native range of brook trout, Salvelinus fontinalis. Canadian Journal of Fisheries and Aquatic Sciences 47, 10651070.
  • Meisner, J. D. (1990b). Potential loss of thermal habitat for brook trout, due to climate warming, in two southern Ontario streams. Transactions of the American Fisheries Society 119, 282291.
  • Meisner, J. D., Rosenfeld, J. S. & Regier, H. A. (1988). The role of groundwater in the impact of climate warming on stream salmonines. Fisheries 13, 27.
  • Mohseni, O., Erickson, D. L. & Stefan, H. G. (1999). Sensitivity of stream temperatures in the United States to air temperatures projected under a global warming scenario. Water Resources Research 35, 37233733.
  • Mohseni, O., Erickson, D. L. & Stefan, H. G. (2002). Upper bounds for stream temperatures in the contiguous United States. Journal of Environmental Engineering 128, 411.
  • Mohseni, O., Stefan, H. G. & Eaton, J. G. (2003). Global warming and potential changes in fish habitat in U.S. streams. Climatic Change 59, 389409.
  • Mugodo, J., Kennard, M., Liston, P., Nichols, S., Linke, S., Norris, R. H. & Lintermans, M. (2006). Local stream habitat variables predicted from catchment scale characteristics are useful for predicting fish distribution. Hydrobiologia 572, 5970.
  • Nakano, S., Kitano, F. & Maekawa, K. (1996). Potential fragmentation and loss of thermal habitats for charrs in the Japanese archipelago due to climatic warming. Freshwater Biology 36, 711722.
  • Nelson, K. C. & Palmer, M. A. (2007). Stream temperature surges under urbanization and climate change: data, models, and responses. Journal of the American Water Resources Association 43, 440452.
  • Nelson, K. C., Palmer, M. A., Pizzuto, J. E., Moglen, G. E., Angermeier, P. L., Hilderbrand, R. H., Dettinger, M. & Hayhoe, K. (2009). Forecasting the combined effects of urbanization and climate change on stream ecosystems: from impacts to management options. Journal of Applied Ecology 46, 154163.
  • Olden, J. D. & Jackson, D. A. (2002). A comparison of statistical approaches for modelling fish species distributions. Freshwater Biology 47, 19761995.
  • Peterson, J. T. & Kwak, T. J. (1999). Modeling the effects of land use and climate change on riverine smallmouth bass. Ecological Applications 9, 13911404.
  • Pilgrim, J. M., Fang, X. & Stefan, H. G. (1998). Stream temperature correlations with air temperatures in Minnesota: implications for climate warming. Journal of the American Water Resources Association 34, 11091121.
  • Rahel, F. J. (2002). Using current biogeographic limits to predict fish distributions following climate change. In Fisheries in a Changing Climate (McGinn, N. A., ed.), pp. 99109. Bethesda, MD: American Fisheries Society.
  • Rahel, F. J., Keleher, C. J. & Anderson, J. L. (1996). Potential habitat loss and populations fragmentation for cold water fish in the North Platte River drainage of the Rocky Mountains: response to climate warming. Limnology and Oceanography 41, 11161123.
  • Schindler, D. W., Augerot, X., Fleishman, E., Mantua, N. J., Riddell, B., Ruckelhaus, M. H., Seeb, J. & Webster, M. (2008). Climate change, ecosystem impacts, and management for Pacific salmon. Fisheries 33, 502506.
  • Sharma, S. & Jackson, D. A. (2008). Predicting smallmouth bass (Micropterus dolomieu) occurrence across North America under climate change: a comparison of statistical approaches. Canadian Journal of Fisheries and Aquatic Sciences 65, 471481.
  • Shuter, B. J. & Post, J. R. (1990). Climate, population viability, and the zoogeography of temperate fishes. Transactions of the American Fisheries Society 119, 314336.
  • Shuter, B. J., Minns, C. K. & Lester, N. (2002). Climate change, freshwater fish, and fisheries: case studies from Ontario and their use in assessing potential impacts. In Fisheries in a Changing Climate (McGinn, N. A., ed.), pp. 7788. Bethesda, MD: American Fisheries Society.
  • Sowa, S. P., Annis, G., Morey, M. E. & Diamond, D. D. (2007). A gap analysis and comprehensive conservation strategy for riverine ecosystems of Missouri. Ecological Monographs 77, 301334.
  • Stanfield, L. W., Gibson, S. F. & Borwick, J. A. (2006). Using a landscape approach to identify the distribution and density patterns of salmonids in Lake Ontario tributaries. In Landscape Influences on Stream Habitats and Biological Assemblages (Hughes, R. M., Wang, L. & Seelbach, P. W., eds) pp. 601622. Bethesda, MD: American Fisheries Society.
  • Steen, P. J., Passino-Reader, D. & Wiley, M. J. (2006). Modeling brook trout presence and absence from landscape variables using four different analytical methods. In Influence of Landscapes on Stream Habitats and Biological Assemblages. (Hughes, R. M., Wang, L. & Seelbach, P. W., eds), pp. 513529. Bethesda, MD: American Fisheries Society.
  • Steen, P. J., Zorn, T. G., Seelbach, P. W. & Schaeffer, J. S. (2008). Classification tree models for predicting distributions of Michigan stream fish from landscape variables. Transactions of the American Fisheries Society 137, 976996.
  • Steen, P. J., Wiley, M. J. & Schaeffer, J. S. (2010). Predicting future changes in Muskegon River watershed game fish distributions under future land cover alteration and climate change scenarios. Transactions of the American Fisheries Society 139, 396412.
  • Stefan, H. G. & Preud’homme, E. B. (1993). Stream temperature estimation from air temperature. Water Resources Bulletin 29, 2745.
  • Steinberg, D. & Colla, P. (1997). CART – Classification and Regression Trees. San Diego, CA: Salford Systems.
  • Stewart, J., Mitro, M., Roehl, E. & Risley, J. (2006). Numerically optimized empirical modeling of highly dynamic, spatially expansive, and behaviorally heterogeneous hydrologic systems, Part 2. Proceedings of the 7th Annual Conference on Hydroinformatics, Nice, France 2006, 18.
  • Turgeon, K. & Rodriguez, M. A. (2005). Predicting microhabitat selection in juvenile Atlantic salmon Salmo salar by the use of logistic regression and classification trees. Freshwater Biology 50, 539551.
  • Van Sickle, J., Baker, J., Herlihy, A., Bayley, P., Gregory, S., Haggerty, P., Ashkenas, L. & Li, J. (2004). Projecting the biological condition of streams under alternative scenarios of human land use. Ecological Applications 14, 368380.
  • Wang, L., Lyons, J., Kanehl, P. & Gatti, R. (1997). Influence of watershed land use on habitat quality and biotic integrity in Wisconsin streams. Fisheries 22, 612.
  • Wang, L., Lyons, J., Kanehl, P., Bannerman, R. & Emmons, E. (2000). Watershed urbanization and changes in fish communities in southeastern Wisconsin streams. Journal of the American Water Resources Association 36, 11731189.
  • Wang, L., Lyons, J., Kanehl, P. & Bannerman, R. (2001). Impacts of urbanization on stream habitat and fish communities across multiple spatial scales. Environmental Management 28, 255266.
  • Wang, L., Lyons, J. & Kanehl, P. (2003). Impacts of urban land cover on trout streams in Wisconsin and Minnesota. Transactions of the American Fisheries Society 132, 825839.
  • Wehrly, K., Wiley, M. J. & Seelbach, P. W. (2003). Classifying regional variation in thermal regime based on stream fish community patterns. Transactions of the American Fisheries Society 132, 1838.
  • Wehrly, K. E., Brenden, T. O. & Wang, L. (2009). A comparison of statistical approaches for predicting stream temperatures across heterogeneous landscapes. Journal of the American Water Resources Association 45, 986997.
  • Whitledge, G. W., Rabeni, C. F., Annis, G. & Sowa, S. P. (2006). Riparian shading and groundwater enhance growth potential for smallmouth bass in Ozark streams. Ecological Applications 16, 14611473.
  • Wiley, M. J., Kohler, S. L. & Seelbach, P. W. (1997). Reconciling landscape and local views of aquatic communities: lessons from Michigan trout streams. Freshwater Biology 37, 133148.
  • Williams, J. E., Haak, A. L., Neville, H. M. & Colyer, W. T. (2009). Potential consequences of climate change to persistence of cutthroat trout populations. North American Journal of Fisheries Management 29, 533548.
  • Winfield, I. J., James, J. B. & Fletcher, J. M. (2008). Northern pike (Esox lucius) in a warming lake: changes in population size and individual condition in relation to prey abundance. Hydrobiologia 601, 2940.
  • Zorn, T., Seelbach, P. W. & Wiley, M. J. (2002). Distributions of stream fishes and their relationship to stream size and hydrology in Michigan's Lower Peninsula. Transactions of the American Fisheries Society 131, 7085.

Electronic References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. References
  8. Electronic References
  • IPCC (2007). Fourth Assessment Report (AR4). Climate Change 2007: The Physical Science Basis. Geneva: IPCC. Available at http://www.ipcc.ch/publications_and_data/publications and data.htm. Accessed May 2010.
  • Lyons, J. (1992). Using the Index of Biotic Integrity (IBI) to Measure Environmental Quality in Warmwater Streams of Wisconsin. St Paul, MN: U.S. Department of Agriculture, Forest Service, North Central Forest Experiment Station. Available at http://www.nrs.fs.fed.us/pubs/213/
  • Roehl, E., Risley, J., Stewart, J. & Mitro, M. (2006). Numerically optimized empirical modeling of highly dynamic, spatially expansive, and behaviorally heterogeneous hydrologic systems, Part 1. In Proceedings of the iEMSs Third Biennial Meeting: Summit on Environmental Modelling and Software (Voinov, A., Jakeman, A. J. & Rizzoli, A. E. eds). International Environmental Modelling and Software Society, Burlington, USA, July 2006. Available at http://www.iemss.org/iemss2006/sessions/all.html. Accessed May 2010.
  • WICCI (2009). Wisconsin Initiative on Climate Change Impacts: change in annual average temperature from 1959 to 2006. Madison, WI: University of Wisconsin. Available at http://ccr.aos.wisc.edu/cwg/. Accessed May 2010.