Introduction

Salmonids and other fishes adapted to cold water are vulnerable to changes in thermal conditions that may be attributable to climate warming. Streams that support trout maintain relatively cold summer maximum water temperatures (Wehrly et al., 2007; Lyons et al., 2010), so warming of stream temperatures from climate change may threaten trout population persistence (Lyons et al., 2010; Mitro et al., 2011; Roberts et al., 2013). The state of Wisconsin in the north-central United States has rich and varied cold-water resources, including > 21,200 km of trout streams that support fisheries for Brook Trout Salvelinus fontinalis and Brown Trout Salmo trutta. Brook Trout are native to Wisconsin, and Brown Trout became naturalized following their introduction to Wisconsin in the late nineteenth century. Wisconsin’s climate has become warmer and wetter since the 1950s, with annual average nighttime low temperatures increasing 0.6–2.2°C, annual average daytime high temperatures increasing 0.3–0.6°C, and average annual precipitation increasing 50–100 mm for 8 km × 8 km grid cells across Wisconsin for 1950–2006 (Kucharik et al., 2010; WICCI, 2011). Identification of fish populations in streams vulnerable to changing climatic conditions has therefore become critical for aiding resource management agencies in the development and implementation of climate-change adaptation strategies (Mitro et al., 2011; WICCI, 2011).

Initial efforts to model fish distribution responses to climate change in Wisconsin began with the development of fish species-distribution models based on classification-tree analyses (Lyons et al., 2010). This modeling related fish species presence and absence to environmental attributes that were associated with stream reaches at multiple scales in a geographic information systems (GIS) framework. Metrics were also included to describe climate, stream flow, and water temperature. Stream-flow variables were estimated from an unpublished regression model referenced by Lyons et al. (2010), and stream water temperatures were estimated from an artificial neural network model described by Roehl et al. (2006) and Stewart et al. (2006). These initial modeling efforts were limited because water temperature was assumed to increase the same in all streams in response to increasing air temperature, and a mechanistic link between changes in precipitation, groundwater recharge, and stream temperature, which could account for differences in vulnerability among streams, was not included.

Lyons et al. (2010) predicted Brook Trout and Brown Trout presence and absence under current climate conditions and projected presence and absence for three future scenarios: limited warming of 1°C (air) and 0.8°C (water), moderate warming of 3°C (air) and 2.4°C (water), and major warming of 5°C (air) and 4°C (water). Significant losses in habitat suitable for trout were projected to occur under all three warming scenarios, with Brook Trout losing 43.6%, 94.4%, and 100% of currently supportive stream length and Brown Trout losing 7.9%, 33.1%, and 88.2% under the three ascending warming scenarios. These models indicated that climate change had the potential to cause major declines in trout distribution in Wisconsin, but limitations of the models suggested that potential declines may have been overstated.

Shortcomings of the models were addressed in subsequent efforts that more realistically characterized changes in stream temperature in response to climate warming. Updated models included a soil–water-balance model integrated with an artificial neural network model to link precipitation to groundwater recharge and stream temperature (Stewart et al., 2015). Updated models also included climate data obtained from the University of Wisconsin Center for Climatic Research. Climate data included air temperature and precipitation projected for the mid-twenty-first century (2046–2065) for the A1B global emissions scenario (moderate to high emissions) using general circulation models (GCMs) downscaled for Wisconsin (Notaro et al., 2011). Herein, we used the latest stream temperature and fish distribution models with climate projections from downscaled GCMs to predict current (late twentieth century) and project future (mid-twenty-first century) Brook Trout and Brown Trout distribution in Wisconsin streams.

Methods

We used a web-based species-distribution model, FishVis (https://ccviewer.wim.usgs.gov/FishVis/#) (Stewart et al., 2016) to predict current and project future Brook Trout and Brown Trout distribution in Wisconsin streams. FishVis is a regional decision support tool developed by the U.S. Geological Survey in cooperation with Michigan State University, Michigan Department of Natural Resources Institute of Fisheries Research, and the Wisconsin Department of Natural Resources. FishVis represents all stream reaches and associated environmental and biological data at the 1:100,000 scale in Minnesota, Wisconsin, Michigan, and New York, and Great Lakes basin portions of Illinois, Indiana, Ohio, and Pennsylvania. Random forest models were developed and included in FishVis to predict the suitability of stream habitat for Brook Trout and Brown Trout and 12 other fish species in individual reaches based on adjacent and upstream channel characteristics, surficial geology, landcover, and climate (Stewart et al., 2016).

Random forest models for Brook Trout and Brown Trout used outputs from models predicting or projecting stream flow and stream temperature, each of which were responsive to changes in air temperature and precipitation. The stream-flow model used regression models to characterize stream-flow exceedance for different seasons and flows. The stream temperature model developed for Wisconsin was an artificial neural network model integrated with a soil–water-balance model that used climate data to predict or project time series of stream temperatures (SWB-ANN model; see Stewart et al., 2015, 2016).

Projected future climate conditions were obtained for the A1B scenario (moderate to high emissions) using general circulation models (GCMs) downscaled for Wisconsin (Notaro et al., 2011). The Wisconsin SWB-ANN model included 10 GCMs (Stewart et al., 2015) and FishVis fish species occurrence models used 13 GCMs (Stewart et al., 2016). Empirical models of Brook Trout and Brown Trout occurrence were based on the fish presence and absence data and environmental characteristics compiled for stream sites surveyed across Wisconsin. Current Brook Trout and Brown Trout presence and absence were predicted using environmental characteristics compiled in a GIS for all 41,097 confluence-to-confluence stream reaches in Wisconsin (83,784 km total, mean reach length 2.0 km, maximum reach length 26.7 km). Future (mid-twenty-first century) Brook Trout and Brown Trout presence and absence were projected by rerunning the models using air temperature and precipitation projections from downscaled GCMs (Stewart et al., 2016).

FishVis was used to predict the probability of occurrence for Brook Trout and Brown Trout under current conditions (late twentieth century; 1961–2000) and to project the probability of occurrence under future conditions (mid-twenty-first century; 2046–2065) for all stream reaches in Wisconsin. Probability of occurrence was defined as the likelihood that a fish species will be present and was predicted by Random Forest models constructed from the observed presence–absence data for fish survey sites and environmental variables for those sites (Stewart et al., 2016). A thermal constraint was also applied so that Brook Trout or Brown Trout were not classified as present if the predicted or projected average maximum daily mean water temperature for a stream reach was warmer than 24.6°C (Lyons et al., 2010; Stewart et al., 2016). A fish species was classified as being present when the probability of occurrence was ≥ 50% and classified as being absent when the probability of occurrence was < 50%. Overall model accuracy was 82.7% for Brook Trout and 82.1% for Brown Trout (Stewart et al., 2016).

The vulnerability of Brook Trout and Brown Trout to changes in climatic conditions was evaluated by comparing changes in probability of occurrence between the late twentieth and mid-twenty-first century periods. Vulnerability or loss of a species from a stream reach was based on the percentage of 13 GCMs that indicated a species currently present would likely be absent in the future. Gain of a species in a stream reach was based on the percentage of 13 GCMs that indicated a species currently absent from a stream reach would likely be present in the future. Sensitivity (loss or gain) of a species in a stream reach was based on the percentage of 13 GCMs that indicated a species would likely change from being absent to present or vice versa.

Results

Brook Trout were predicted to be present in 16,883 stream reaches or 34,251 stream km and absent in 24,214 reaches or 49,533 km in Wisconsin in the late twentieth century (Fig. 1a) and projected to be present in 5,515 reaches or 10,995 km and absent in 35,582 reaches or 72,789 km in the mid-twenty-first century (Fig. 1b). The projected loss of Brook Trout-stream habitat between the late twentieth and mid-twenty-first century periods was 68%.

Fig. 1
figure 1

Brook Trout probability of occurrence for current (a; late twentieth century) and future (b; mid-twenty-first century) periods. Colors indicate categories of probability of occurrence: red (0–0.2), orange (0.21–0.4), yellow (0.41–0.6), light green (0.61–0.8), and dark green (0.81–1)

Full size image

Brown Trout were less prevalent than Brook Trout in the current period, but were projected to lose less stream habitat in the future than Brook Trout. Brown Trout were predicted to be present in 10,392 stream reaches or 20,011 stream km and absent in 30,705 reaches or 63,774 km in Wisconsin in the late twentieth century (Fig. 2a) and projected to be present in 7,069 reaches or 13,668 km and absent in 34,028 reaches or 70,117 km in the mid-twenty-first century (Fig. 2b). The projected loss of Brown Trout-stream habitat between the late twentieth and mid-twenty-first century periods was 32%.

Fig. 2
figure 2

Brown Trout probability of occurrence for current (a; late twentieth century) and future (b; mid-twenty-first century) periods. Colors indicate categories of probability of occurrence: red (0–0.2), orange (0.21–0.4), yellow (0.41–0.6), light green (0.61–0.8), and dark green (0.81–1)

Full size image

Trout-stream habitat was concentrated in the western and northern parts of Wisconsin, with a narrow band connecting southwestern and northeastern Wisconsin in the late twentieth century (Figs. 1a and 2a). Projected loss of trout habitat in the mid-twenty-first century occurred throughout the state, with higher concentrations of remaining trout-stream habitat in the Driftless Area of western Wisconsin and the Lake Superior basin to the north (Figs. 1b and 2b). On a finer-scale, for example, the probability of occurrence decreased within several counties in northeastern Wisconsin for Brook Trout (Fig. 3) and several counties in the Driftless Area of western Wisconsin for Brown Trout (Fig. 4).

Fig. 3
figure 3

Brook Trout probability of occurrence for current (a; late twentieth century) and future (b; mid-twenty-first century) periods for northeastern Wisconsin including parts of Florence, Forest, Langlade, Oconto, and Marinette counties. Colors indicate categories of probability of occurrence: red (0–0.2), orange (0.21–0.4), yellow (0.41–0.6), light green (0.61–0.8), and dark green (0.81–1)

Full size image
Fig. 4
figure 4

Brown Trout probability of occurrence for current (a; late twentieth century) and future (b; mid-twenty-first century) periods for southwestern Wisconsin including La Crosse, Monroe, and Vernon counties. Colors indicate categories of probability of occurrence: red (0–0.2), orange (0.21–0.4), yellow (0.41–0.6), light green (0.61–0.8), and dark green (0.81–1)

Full size image

Stream reaches where Brook Trout or Brown Trout were predicted to be present in the late twentieth century (Figs. 5a and 6a) were vulnerable to loss of trout to varying degrees depending on the percent of GCMs that indicated loss in the mid-twenty-first century (Table 1; Figs. 5b and 6b). Brook Trout were more vulnerable to loss of habitat than Brown Trout (Table 1). The opportunity to gain stream habitat (Table 2; Figs. 5c and 6c) was minimal compared to the vulnerability to lose stream habitat, and Brown Trout had more opportunity to gain habitat than Brook Trout (Table 2). Sensitivity to loss or gain of habitat was also greater for Brook Trout than Brown Trout (Table 3; Figs. 5d and 6d).

Fig. 5
figure 5

a Predicted Brook Trout occurrence (black) and absence (gray) for the current (late twentieth century) period; b Brook Trout vulnerability to habitat loss (percent of GCMs that project loss) for the future (mid-twenty-first century) period. Colors indicate the percent of GCMs that project Brook Trout loss: gray = already absent in the late twentieth century period, green = 0% (not vulnerable and likely to remain present), and light red = 1–20% (less vulnerable to loss) to dark red = 81–100% (more vulnerable to loss); c Brook Trout opportunity to gain habitat (percent of GCMs that project gain) for the future (mid-twenty-first century) period. Colors indicate the percent of GCMs that project Brook Trout gain: green = already present in the late twentieth century period, gray = 0% (no opportunity to gain and likely to remain absent), and light blue = 1–20% (less opportunity to gain) to dark blue = 81–100% (more opportunity to gain); d Brook Trout sensitivity to loss or gain of habitat (percent of GCMs that project loss or gain) for the future (mid-twenty-first century) period. Colors indicate the percent of GCMs that project Brook Trout loss or gain: green = present in the late twentieth century period and not likely to disappear, gray = absent in the late twentieth century period and not likely to reappear, and light purple = 1–20% (less sensitive to loss or gain) to dark purple = 81–100% (more sensitive to loss or gain)

Full size image
Fig. 6
figure 6

a Predicted Brown Trout occurrence (black) and the absence (gray) for the current (late twentieth century) period; b Brown Trout vulnerability to habitat loss (percent of GCMs that project loss) for the future (mid-twenty-first century) period. Colors indicate the percent of GCMs that project Brown Trout loss: gray = already absent in the late twentieth century period, green = 0% (not vulnerable and likely to remain present), and light red = 1–20% (less vulnerable to loss) to dark red = 81–100% (more vulnerable to loss); c Brown Trout opportunity to gain habitat (percent of GCMs that project gain) for the future (mid-twenty-first century) period. Colors indicate the percent of GCMs that project Brown Trout gain: green = already present in the late twentieth century period, gray = 0% (no opportunity to gain and likely to remain absent), and light blue = 1–20% (less opportunity to gain) to dark blue = 81–100% (more opportunity to gain); d Brown Trout sensitivity to loss or gain of habitat (percent of GCMs that project loss or gain) for the future (mid-twenty-first century) period. Colors indicate the percent of GCMs that project Brown Trout loss or gain: green = present in the late twentieth century period and not likely to disappear, gray = absent in the late twentieth century period and not likely to reappear, and light purple = 1–20% (less sensitive to loss or gain) to dark purple = 81–100% (more sensitive to loss or gain)

Full size image
Table 1 Vulnerability to stream-habitat loss for Brook Trout and Brown Trout in the mid-twenty-first century based on projections of 13 general circulation models (GCMs) and the A1B global emissions scenario
Full size table
Table 2 Opportunity to gain stream habitat for Brook Trout and Brown Trout in the mid-twenty-first century based on projections of 13 general circulation models (GCMs) and the A1B global emissions scenario
Full size table
Table 3 Sensitivity to loss or gain of stream habitat for Brook Trout and Brown Trout in the mid-twenty-first century
Full size table

Discussion

Our results indicated that changes in stream habitat associated with climate change could lead to declines in the occurrence and distribution of Brook Trout and Brown Trout in Wisconsin streams. Projected losses occurred throughout the current distribution of trout in Wisconsin, but streams in some parts of the state were more resistant or resilient to climate warming than others. The Driftless Area of western-southwestern Wisconsin, with its karst topography that provides abundant cold groundwater to streams, and the Lake Superior basin, with the coldest summer temperatures in the state, were less vulnerable to cold water habitat loss (Stewart et al., 2015).

Groundwater temperature will change slower than surface-water temperature in response to warming air temperature, and increased precipitation can improve groundwater recharge and baseflow in streams (WICCI, 2011). An increase in baseflow in Driftless Area streams beginning around 1970 coincided with increased precipitation and improvements in agricultural land management (Juckem et al., 2008), which likely enabled successful recolonization or establishment of stocked trout populations in streams that had lost trout because of degraded stream conditions (Hoxmeier et al., 2015). Cold-water refugia will persist for the foreseeable future, which will allow cold-water fishes like Brook Trout and Brown Trout to persist, albeit in fewer stream reaches (Isaak et al., 2016; Merriam et al., 2017).

Our models for Brook Trout and Brown Trout were independent of one another and did not consider competition between the two species where they co-occur. Both species were predicted or projected to occur in the same stream for some streams, and some streams were projected to become unsuitable for one species but become suitable for the other. Brook Trout and Brown Trout share similar thermal tolerance limits (Wehrly et al., 2007), but within thermal tolerance limits are a series of decreasing preferred and optimal temperature ranges for functions such as feeding and growth. Brook Trout have an adaptation to and preference for colder water within thermal tolerance limits compared to Brown Trout (Behnke, 2002), which may be why Brown Trout were shown by the models to have more opportunity than Brook Trout to gain habitat.

Our results confirm others that projected significant declines in suitable habitat for trout. For example, habitat supporting nonnative Brook Trout declined in the inland western United States, with projected losses of 44% of its range in the 2040s and 77% in 2080s under the A1B emissions scenario (Wenger et al., 2011). Habitat supporting Brown Trout also declined less than habitat supporting Brook Trout and lost 16% and 48% of their range for the two periods (Wenger et al., 2011). A study in the Aragon River basin in northern Spain, where Brown Trout are native, similarly showed that Brown Trout may lose 50% of current suitable habitat (Almodovar et al., 2012).

Climate-related changes in stream temperature and flow are expected to alter trout distribution at the landscape level (Kovach et al., 2016, Santiago et al., 2017). Loss of salmonids may ultimately be caused by temperatures elevated beyond thermal tolerance limits (Wehrly et al., 2007). However, changes in temperature that cause changes in how aquatic organisms interact may lead to climate-related losses at lower than lethal temperatures (Cahill et al., 2013). For example, species interactions, in which an ectoparasitic copepod Salmincola edwardsii infects the gills of Brook Trout but not co-occurring Brown Trout, may lead to Brook Trout loss under elevated thermal and lowered stream-flow conditions that favor the ectoparasite’s life cycle (Mitro, 2016).

FishVis will be useful to fisheries managers to support resource allocation decisions for building resiliency to climate change in cold-water streams. Wisconsin’s trout streams have rebounded in recent decades from a century of poor land-use practices that degraded thermal and physical stream habitat. Land, riparian, and water management strategies, including stream restoration, are being used to increase trout fisheries and conserve cold thermal habitat. Two types of adaptation strategies can help to protect cold-water stream fisheries threatened by climate change. The first concerns how management efforts are directed, and the second concerns management activities that can potentially offset negative effects of climate warming.

We recommend a triage approach for examining potential effects of climate warming on trout populations in streams. FishVis can be used to identify streams that may experience losses of cold-water fishes, even with intensive management, streams that may already be buffered to changes in climate and are likely to persist as trout habitat in the absence of management, and streams that could be maintained as viable trout habitat by implementing adaptation strategies. We think management activities could be directed to streams where trout population persistence depends most on management intervention.

Adaptation strategies that can potentially offset negative effects of climate warming can be implemented across scales from the landscape to the stream. At the landscape level, strategies may include improved agricultural land-use practices such as no-till farming, contour plowing, rotational grazing, use of cover crops during winter, and establishment of riparian buffers (Lyons et al., 2000; Blann et al., 2002). Enrollment of the most environmentally sensitive lands into protective conservation programs and limiting impervious surfaces may also help increase groundwater recharge to maintain cold stream temperatures (Wang et al., 2003; Marshall et al., 2008). Managing riparian vegetation to provide shading may also help maintain cold thermal conditions suitable for trout (Cross et al., 2013). Instream strategies may include sloping erosive streambanks to open streams to their floodplain and to reduce sediment loading, narrowing and deepening stream channels to maintain cold stream temperatures, and installing physical habitat to promote self-sustaining trout populations (Hunt, 1976). Used in combination, landscape-conservation practices and triaging instream habitat-restoration efforts can potentially buffer vulnerable streams to some of the effects of climate warming.