Walleye (Sander vitreus) populations in Wisconsin are near the center of their geographical range and support valuable fisheries. The role of seasonal temperature and precipitation in Walleye recruitment was examined using regression tree analysis. Climatological variables were estimated at the 8 km2 scale and Walleye recruitment was estimated based on 298 individual electrofishing surveys. Estimated changes in Walleye recruitment between 1950 and 2006 were examined based on changes in explanatory climatological variables. Spring precipitation and summer maximum temperature were significant predictors of age-0 Walleye density and mean estimated changes in these variables between 1950 and 2006 were used to estimate changes in Walleye recruitment. The model predicted a small overall increase in Walleye recruitment and provides insight into the direct role of climatological variables in Walleye recruitment. However, given the low explanatory power (R2 = 0.103), it is likely that historic climatological changes have had a limited effect on recruitment levels.
Given historical changes in climatological variables, there has likely been and will continue to be distribution shifts for various species. Fish populations may be extirpated from some areas and expand into others (Chu et al., 2005; Lyons et al., 2010) as climatological changes cause suitable habitat to become more or less abundant. Wisconsin is approximately in the center of the current spatial distribution of Walleye (Sander vitreus) which ranges from the Northwest Territories, Canada, south to New Mexico, United States, in the West and from northern Quebec, Canada, to Alabama, United States, in the east (Becker, 1983). Expected range contraction and expansion primarily occurs at the edges of current distributions so it is unlikely that Walleye will be extirpated from the waters of Wisconsin in the foreseeable future. However, other important aspects of fish populations such as growth (Brandt et al., 2002) and recruitment (Casselman, 2002; MacKenzie and Koster, 2004) may be altered in response to changes in climatological variables. Walleye population metrics of interest may increase or decrease in particular areas of the state if climatological variables change in a fashion that is more or less favorable.
Walleye recruitment strength has been shown to be related to a number of abiotic variables in lakes and reservoirs. Increasing or relatively high water levels are positively related to Walleye recruitment in some reservoirs (Quist et al., 2003) and lakes (Chevalier, 1977; Nelson and Walburg, 1977) with the suggestion that higher water levels increase spawning habitat availability. Spring water temperatures also influence Walleye recruitment. Increased spring water temperatures are associated with greater year class strength which may result from a number of mechanisms including increased growth, decreased risk to predation owing to the shortened development time, and increased exposure to the effects of wind (Quist et al., 2003). In addition, the behavior of spring temperatures has been associated with Walleye recruitment. Faster spring warming rates have been associated with increased recruitment in Lake Erie, OH (Madenjian et al., 1996) and decreased variation in May water temperatures lead to increased recruitment in Escanaba Lake, WI (Hansen et al., 1998). Hansen et al. (1998) proposed direct mortality, decreased growth, decreased prey availability, decreased spawning activity, and delayed embryological development as potential mechanisms for the inverse relationship between variation in May water temperatures and Walleye recruitment. Changes in these variables or others over time may affect Walleye recruitment in a positive or negative fashion.
Climatological variables have not changed in a uniform manner throughout the world, region, state, or even smaller units of area (IPCC, 2007). In fact, the changes have likely varied substantially in areas even in close proximity. Kucharik et al. (2010) used methods developed in Serbin and Kucharik (2009) to generally demonstrate higher maximum and minimum temperatures as well as increased precipitation in Wisconsin USA but also found substantial heterogeneity in climatological change from 1950–2006 at the resolution of 8 km2. Thus, it is unlikely that any trends in Walleye recruitment related to climatological variables will be uniform throughout the state. Understanding where Walleye recruitment may have been impeded or enhanced will help dictate future expectations and management actions such as stocking or harvest limitation.
Walleye populations in Wisconsin support both recreational and tribal subsistence fisheries. The Walleye is the most commonly targeted fish species in Wisconsin recreational fisheries (McClanahan, 2003) and six Chippewa tribal bands participate in annual Walleye harvest that occurs primarily during a spring spear fishing effort in the northern third of the state (U.S. Department of the Interior, 1991). Given the importance of Walleye to both recreational anglers and tribal interests, it is helpful to understand the role of climate in Walleye recruitment as climate variables have trended over time (IPCC, 2007) in order to adjust management actions accordingly.
The majority of Wisconsin's Walleye fisheries occur in the northern third of Wisconsin and the current management system appears to result in sustainable Walleye populations (Beard et al., 2003). There have been no significant overall downward trends in adult Walleye density (Cichosz, 2010). However, if localized climate changes have occurred that resulted in increased Walleye recruitment in some areas and decreased Walleye recruitment in others, localized trends within northern Wisconsin may be masked.
Our objectives were to develop a model to predict age-0 Walleye density based on seasonal temperature and precipitation and to determine whether Walleye recruitment has been altered due to changes in climate over the past half century. Age-0 Walleye abundance is of interest because it can be linked to adult density (Johnson, 1999) and the survival of juvenile Walleyes from spawning to their first fall have been linked to climatological conditions (Chevalier, 1977; Nelson and Walburg, 1977; Madenjian et al., 1996; Hansen et al., 1998; Quist et al., 2003). Changes in Walleye recruitment associated with changes in climate were predicted by developing an explanatory model for Walleye recruitment in Wisconsin based on recent climatological data and then expected Walleye recruitment in 1950 and 2006 was predicted using this model.
A model using data collected between 1997 and 2006 to predict juvenile Walleye recruitment was developed based on climatological variables that occurred from hatching to their first fall and then this model was used to predict recruitment in 1950 and 2006. This time period was chosen because Kucharik et al. (2010) provided relatively high resolution (8 km2) estimates of changes in climatological variables between 1950 and 2006. An explanatory model with juvenile Walleye density as the response variable and the mean values of seasonal climatological variables (maximum temperature, minimum temperature, and precipitation) as explanatory variables was developed. Seasons were defined as follows: spring = March, April and May; summer = June, July and August; fall = September, October and November; and winter = December, January and February in the year prior to hatching.
where R is the number of age-0 Walleyes per hectare, T is temperature in degrees Celsius, and N is the number of age-0 Walleye caught per kilometer of electrofishing.
Monthly downscaled climatological data from Serbin and Kucharik (2009) were used to develop our explanatory model for Walleye recruitment. Briefly, Serbin and Kucharik (2009) used historical records from 1950–2006 from 176 climate stations located throughout the state of Wisconsin to downscale climatological conditions to a resolution of 8 km2 using an inverse distance weighting algorithm. Using ArcMap 9.3 (ESRI, 2008), a spatial join between the coordinates (latitude and longitude) of the lake where the age-0 Walleye survey occurred and the closest downscaled estimated climate data grid cell was performed, linking the climate data to the estimated Walleye age-0 densities. Only climatological variables experienced by Walleyes from hatching to their first fall were included. These data included spring maximum temperature, spring minimum temperature, spring precipitation, summer maximum temperature, summer minimum temperature, summer precipitation, fall maximum temperature, fall minimum temperature, fall precipitation, and winter precipitation of the year prior to hatching. Winter precipitation was included because it could influence lake levels and amount of runoff experienced by newly hatched Walleye fry in the spring.
A regression tree approach was used to develop the model to predict age-0 Walleye density based on climatological conditions. The R package “rpart” was used to develop the regression tree (R Core Development Team, 2005). Unlike linear regression models, regression trees inherently detect interactions between independent variables and normality of predictor values is not required (Quinn and Keough, 2006). Regression trees begin with all the data and split them into smaller nodes as long as there is a predefined reduction in the amount of unexplained variation. Any split that does not decrease the overall lack of fit by a predetermined amount is not attempted. Splits that did not decrease the lack of fit by setting the “complexity parameter” (cp) in rpart equal to 0.05 were not attempted. Other criteria for the formation of a node in the analysis were the minimum number of observations in the terminal node (leaf) and the minimum number in a node to attempt an additional split. Default values in rpart for these criteria were used which were 20 for an attempted split (“minsplit” = 20) and 7 for the number in each terminal node (“minbucket” = “minsplit”/3). Finally, method was set to “anova,” which is appropriate if the response variable is continuous. In rpart the R2 is equivalent to 1-relative error (Therneau and Atkinson, 1997).
In order to determine how changes in climate may have altered Walleye recruitment in Wisconsin, predicted downscaled average climatological conditions in 1950 from each 8 km2 grid cell obtained from Kucharik et al. (2010) were incorporated into the explanatory regression tree model developed with data obtained from 1997–2006. Kucharik et al. (2010) used linear regression to estimate the change in climatological variables from 1950–2006 for each 8 km2 grid cell in Wisconsin (as determined by Serbin and Kucharik, 2009). A spatial join in ArcMap 9.3 (ESRI, 2008) of all lakes and flowages in the Wisconsin DNR 1:24,000 scale hydrography layer (version 6) in the state of Wisconsin greater or equal to 20 hectares (as Walleye are generally absent from very small bodies of water [Becker, 1983]) was performed and linked the waters to the closest climatological grid cell. The mean estimated values of seasonal climatological variables in 1950 and 2006 based on the slope and intercept of the regression equations developed by Kucharik et al. (2010) were used to estimate associated changes in Walleye recruitment over that time period throughout the state. These values are not actual measurements from 1950 and 2006 but simply average conditions that could be expected given estimated changes over time. The results of expected Walleye recruitment in 1950 was used to summarize total recruitment in terms of spatial distribution and statewide totals compared to those expected given estimated climate conditions present in 2006. Using acreages of the waters present in Wisconsin, estimated recruitment per hectare was simply multiplied by the number of hectares of lake area for each grid cell based on average climatological changes between 1950 and 2006 to estimate changes in statewide Walleye recruitment.
The regression tree model suggests that spring precipitation and summer maximum temperature affect the abundance of age-0 Walleyes (Figure 2). Total spring precipitation less than 253.8 mm resulted in low age-0 Walleye density regardless of the summer minimum temperature. However, if spring precipitation was greater than 253.8 mm, age-0 Walleye densities were further increased by summers with higher maximum temperatures (Figure 2). The relative error for this model was 0.897 which equates to an R2 value of 0.103 (Therneau and Atkinson, 1997).
Based on localized changes in spring precipitation and summer maximum temperature for 8 km2 grid cells between 1950 and 2006 it is expected that Wisconsin has experienced an overall increase in age-0 Walleye density of approximately 3% between 1950 and 2006. This increase was based on the climatological changes experienced by all lakes greater than 20 ha in Wisconsin based on our assumption that all lakes greater than 20 ha contain Walleyes. However, the expected changes in age-0 Walleye density were spatially heterogeneous with predicted increases the south central part of Wisconsin, while there were no predicted changes in age-0 Walleye density in the remainder of the state (Figure 3). Approximately 33% of the total area of waters presumed to contain Walleye in Wisconsin is located south of the 44 degrees latitude (Figure 3). When considering only waters in southern Wisconsin, Walleye recruitment was predicted to increase by 9.9% between 1950 and 2006. The increases were driven by the predicted increase in spring precipitation in the south central part of Wisconsin (Kucharik et al., 2010). Of the total number of lakes greater than 20 ha in Wisconsin (2,042) only a very small number (27) were predicted to experience increased Walleye recruitment based on our explanatory model (Table 1).
There was an estimated overall increase of 3% in the expected amount of Walleye recruitment associated with changes in climate in Wisconsin between 1950 and 2006. This result agrees with the assessment that overall Walleye populations in Wisconsin have not increased or decreased dramatically in the past 20 years (Cichosz, 2010) and that the current management system is sustainable (Beard et al., 2003). However, it is doubtful that an overall increase in Walleye recruitment of 3% spread over the entire state and 50 years, even if perfectly accurate, could be detected by the public or natural resources managers due to the large number of small lakes, the long time period involved (Pauly, 1995), and the small magnitude of the change.
The climatological variables that were included are biologically reasonable and agree with other investigations into Walleye recruitment. In the model, higher spring precipitation levels were associated with increased Walleye recruitment. Other studies have suggested that increased water levels were associated with increased Walleye recruitment (Nelson and Walburg, 1977; Chevalier, 1977; Quist et al., 2003). Both flow and water levels would be higher in an environment that experienced increased precipitation in most cases. However, while the model suggests that increased spring precipitation is associated with higher Walleye recruitment, this model is based on historic precipitation patterns. Recent precipitation events have been more intense than in the past (Kucharik et al., 2010) and this trend may continue. Since very high flow events may actually act to decrease Walleye recruitment (Quist et al., 2003) high levels of precipitation may not act to increase Walleye recruitment if the events in the future are more intense than they have been in the past as is expected to occur (Kucharik et al., 2010). The model also included summer maximum temperature, with increased maximum temperature being associated with higher Walleye recruitment. Increased Walleye growth has been associated with higher water temperatures in summer in Wisconsin (Serns, 1982; Staggs and Otis, 1996) and increased growth has been associated with higher survival of juvenile Walleyes (Forney, 1976). Higher summer temperatures may result in increased growth and survival of juvenile Walleyes and thereby increased recruitment. The optimal temperature for growth of Walleyes is approximately 26°C (Hokanson and Koenst, 1986) and the model suggested that maximum air temperatures above 27.9°C were associated with increased recruitment. Since water temperature is determined by a combination of maximum and minimum temperatures, a maximum air temperature of 27.9°C may correspond with the near optimal temperature for Walleye growth which in turn may be linked to increased survival. However, the upper lethal temperature for Walleye is approximately 34°C (Hokanson and Koenst, 1986) and although modest increases in summer temperature may act to increase age-0 Walleye density, dramatic increases in summer temperature may act to decrease it.
The explanatory power of our model was very low. Therefore, it is not recommend that lake or area specific management actions be based on our results. However, models with low explanatory value can be useful and can at least set the framework for further research to explain observed changes in Walleye recruitment over time. Since the climatological variables examined seem to explain only a small portion of the observed variance in age-0 Walleye density, other factors and interactions of other factors must play a substantial role in determining year class strength. There are many other chemical, morphometric and biological factors that affect Walleye recruitment and these almost certainly interact with changes in climate making lake specific predictions difficult. In some systems biotic interactions can have an overriding influence on Walleye recruitment (Quist et al., 2003). Interactions related to competitor and predator abundance can affect Walleye recruitment (Forney, 1976; Hansen et al., 1998; Quist et al., 2003; Nate et al., 2003; Fayram et al., 2005). Other species in the community have likely been affected by changes in climatological variables as well, thereby confounding direct predictions of the effects of climate change on age-0 Walleye density.
There were spatially heterogeneous predicted responses of Walleye recruitment based on differential changes in climatological variables. The south central part of Wisconsin is expected to have had modestly increased recruitment while there were no predicted changes in the rest of the state. There have been reported reductions in Walleye recruitment primarily in the northwest part of Wisconsin (WDNR, Thomas Cichosz, WDNR Bureau of Fisheries Management, Madison, WI, pers. communication). These observations are consistent with our findings particularly given the low explanatory power of our model and the fact that other abiotic and biotic factors certainly come into play with regard to Walleye recruitment.
Our results suggest that spatially heterogeneous changes in climatological variables may affect Walleye recruitment differently in different waterbodies in Wisconsin. Therefore, a statewide approach to management may be unwarranted although individual lake management is logistically infeasible and as such an intermediate scale of management may be appropriate. Given the low explanatory power of our model, we recommend further research focused on characterizing climate related changes in biological, physical, and chemical variables that may affect Walleye which themselves may have changed in response to changes in climatological variables.
We thank Chris Kucharik, Michael Notaro, and Dan Vimont for providing climatological data and for helpful discussions regarding this project. In addition, we thank the members of the Wisconsin Initiative on Climate Change Impacts (WICCI) for supporting the work of developing adaptive strategies for climate change impacts in Wisconsin. This project would not have been possible without the outstanding efforts of Wisconsin Department of Natural Resources and Great Lakes Indian Fish and Wildlife Commission fisheries biologists and technicians who collected the Walleye data. Jon Hansen provided a helpful review of an earlier version of this article.
Federal Aid in Sportfish Restoration provided partial funding for this project.