Growth of rainbow trout (Oncorhynchus mykiss) in warm-temperate lakes: implications 1 for environmental change 2 3 Jennifer M. Blair*1,5, Ilia Ostrovsky2, Brendan J. Hicks1, Robert J. Pitkethley3, Paul Scholes4 4 5 1Department of Biological Sciences, Environmental Research Institute, University of 6 Waikato, Private Bag 3105, Hamilton 3240, New Zealand (jennifermblair@gmail.com; 7 b.hicks@waikato.ac.nz) 8 2Israel Oceanographic & Limnological Research, Kinneret Limnological Laboratory, P.O.B. 9 447, Migdal 14950, Israel (ostrovsky@ocean.org.il) 10 3Eastern Region Fish & Game, Private Bag 3010, Rotorua, New Zealand (rob@erfgc.co.nz) 11 4Bay of Plenty Regional Council, PO Box 364, Whakatane 3158, New Zealand 12 (paul.scholes@envbop.govt.nz) 13 5Present address: Kessels & Associates, PO Box 4225, Hamilton 3274, New Zealand 14 15 *Corresponding author (email: jennifermblair@gmail.com, phone: + 64 7 981 1418, fax: +64 16 7 856 0468). 17

Canadian Journal of Fisheries and Aquatic Sciences, 10.1139/cjfas-2012-0409 18

19 20

Abstract 21

To examine the effects of climate and anthropogenic impacts on fish growth, we compared 22 growth rates of rainbow trout (Oncorhynchus mykiss) in nine closely located warm-temperate 23 lakes of contrasting morphometry, stratification and mixing regime, and trophic state. 24 Analyses of long-term mark-recapture data showed that in deep oligotrophic and mesotrophic 25 lakes, trout growth rates increased with increasing indices of lake productivity. In contrast, in 26 shallow eutrophic lakes, where fish habitat volume is constrained by temperature and 27 dissolved oxygen, trout growth rates declined with increasing productivity. Growth rates 28 were higher in lakes with greater volumes of favourable habitat (i.e. dissolved oxygen >6.0 29 mg L-1 and temperature <21°C), and lower in lakes with increased turbidity, chlorophyll a, 30 and nitrogen concentrations. Our findings suggest that lake increases in lake productivity and 31 temperatures as a result of global climatic change are likely to be more detrimental to 32 salmonid habitat quality and volume in shallower, productive lakes, while salmonids will 33 better endure such changes in deeper, oligotrophic lakes. Fishery managers can use this 34 information to aid future stocking decisions for salmonid fisheries in warm-temperate 35 climates. 36 37

Introduction 38

The health and welfare of fish can be influenced by different environmental factors such as 39 water quality parameters, availability of food, volume of habitat, and natural or stocking 40 densities. Growth rate is an important biological parameter that is a key determinant of 41 production and reflects the suitability of ecological conditions for particular fish species 42 (Haddix and Budy 2005, Wells et al. 2008). Therefore, growth rate can be used as a marker 43 for fish population health in specific ecosystems. To help predict fish growth in different 44 conditions, relationships have been derived between fish growth rates or production and 45 various physical, chemical, and biological variables, such as lake depth, nutrient loading, and 46

lake productivity (e.g. Oglesby 1977, Leach et al. 1987, Downing and Plante 1993). These 47 relationships are context- specific. For example, in boreal lakes of British Columbia and 48 Alaska, salmonid survival and size increased with addition of nutrients (Hyatt et al. 2004). 49 However, excessive primary productivity can become detrimental to certain fish species. 50 Foraging of visual predators may be impaired by high turbidity in eutrophic lakes 51 (Beauchamp et al. 1999, Ranåker et al. 2012), and the volume of cool habitat in the 52 hypolimnion may be reduced by low dissolved oxygen (DO) concentrations (Plumb and 53 Blanchfield 2009). In warm, productive lakes, growth rate potential of coolwater species 54 decreases with increasing chl a concentrations (Budy et al. 2011). 55 Lake morphometry influences many variables that are important for fish habitat, e.g. 56 mixing regime, nutrient cycling, temperature, DO, and sediment resuspension (Fee 1979, 57 Genkai-Kato and Carpenter 2005, Håkanson 2005), and can therefore determine habitat 58 conditions for fish populations. The availability of suitable thermal habitat affects total 59 sustained yield of fish in temperate lakes (Christie and Regier 1988), and in salmonids, high 60 temperatures may affect several physiological processes detrimentally, including feeding, 61 growth, and reproduction (Myrick and Cech 2000, Pankhurst and King 2010, Robinson et al. 62 2010). Fish can avoid detrimental warm temperatures in thermally stratified lakes by 63 occupying the cooler hypolimnion, but only if the oxygen regime is favourable (Budy et al. 64 2009). However, shallow lakes usually lack such refuge habitat. Additionally, low 65 hypolimnetic dissolved oxygen (DO) concentrations will reduce the amount of available 66 habitat in thermally stratified lakes (Coutant 1985, Plumb and Blanchfield 2009). 67 Hypolimnetic deoxygenation is a problem not only in monomictic lakes: stratification events 68 in shallow, polymictic, eutrophic lakes can also result in rapid hypolimnetic DO loss (Burger 69 et al. 2007), shrinking fish habitat. 70

Climate warming is predicted to increase lake temperature and duration of 71 stratification, which will reduce habitat for coldwater and coolwater fish in shallow and 72 moderately deep temperate lakes (e.g. in North America; Schindler 2001, Fang et al. 2004). 73 Under these conditions, continuing anthropogenic nutrient supply from catchments and 74 increased internal loads of nutrients from bottom sediments (due to higher temperature and 75 anoxic conditions) will drive enhanced eutrophication and productivity of lakes (see e.g. 76 Jeppesen et al. 2009, Trolle et al. 2011). 77

Several studies have examined the influences of environmental factors on salmonid 78 growth in cool-temperate, continental climates (Plante and Downing 1993, Hyatt et al. 2004). 79 The interacting effects of increased temperature, eutrophication, reduced DO, and decreased 80 habitat volume can make it difficult to understand the mechanisms controlling salmonid 81 growth rates in lakes. Because of likely future temperature increases, it is important to study 82 constraints on salmonid growth in near-stress situations in warmer climates. This will allow 83 us to predict the future effects of multiple stressors, including higher temperatures and trophic 84 state, on fish growth. 85

In this study, we aim to determine the influence of physical habitat (e.g. DO and 86 temperature) and primary production (e.g. nutrient and chl a concentrations) on growth rates 87 and body condition of rainbow trout (Oncorhynchus mykiss) in warm-temperate lakes. This 88 is important because future management and conservation of salmonid populations will need 89 to take into account changes in climate and increases in anthropogenic effects. Our suite of 90 lakes in New Zealand is ideal for this purpose because they are near the geographical limit of 91 rainbow trout distribution, and therefore represent near-stress conditions. The study lakes are 92 geographically close and experience the same climate, but display a range of morphometries, 93 trophic states, and mixing regimes. This gives us a unique opportunity to compare fish 94 growth under a range of conditions in natural systems. 95

We examine growth rates of hatchery-raised rainbow trout from a single population 96 released into these different lakes; this allows us to test the effects of various environmental 97 factors on growth rates while controlling for genetic influences (sensu Reed et al. 2010). We 98 also consider catchment land use, which affects productivity and water quality of lakes by 99 influencing the allochthonous supply of nutrients and organic carbon. We hypothesise that in 100 deeper, cooler, monomictic lakes, moderate nutrient supply may be favourable for rainbow 101 trout growth, while in shallower, warmer, polymictic lakes, additional external or internal 102 supply of nutrients and organic material may negatively affect habitat quality and thus create 103 unfavourable conditions for trout growth. This study will provide information on priorities 104 for lake ecosystem management to sustain salmonid fisheries in warm-temperate lakes in the 105 face of changing climate and lake productivity, leading to better informed management 106 decisions involving species of high social and economic importance. 107

108

Methods 109

Lakes, fish populations, and field sampling 110 The Rotorua area in the Bay of Plenty region of New Zealand’s North Island contains several 111 lakes of varying size, depth, geological origin, and catchment land use; these lakes display a 112 wide range of temperatures and trophic states (Table 1) despite their close geographic 113 proximity (Fig. S1). Eastern Region Fish and Game (ERFG) stocked rainbow trout into nine 114 study lakes that varied in size and trophic state (Fig. S1, Table 1) in austral spring 115 (September) and autumn (May) from a hatchery facility at Ngongotaha, Rotorua. All hatchery 116 parent stocks originated from the Te Wairoa Stream, a tributary of Lake Tarawera. Stocking 117 rates varied due to differences in management strategies and wild spawning populations 118 (Table 2). Trout released in autumn were approximately 1 year old upon release (fork length 119 = 160 ± 3 mm; mean ± SE), and those released in spring were 1.25 years old (190 ± 2 mm). 120 Trout lengths at release did not differ among lakes in either spring (ANOVA; F1,8 = 0.45, P = 121 0.89, n = 9) or autumn (F1,3 = 0.22, P = 0.88, n = 9). 122

We calculated growth rates of rainbow trout in different lakes using data collected 123 from tagged rainbow trout that were recaptured by anglers. We tagged 500 fish from each 124 stocking with batch-coded T-bar anchor tags inserted at the base of the dorsal fin, and 125 measured the fork length of a subsample of fish (mean n = 80) to the nearest 10 mm prior to 126 release (Table 2). Subsequently, ERFG staff measured some trout during creel surveys; 127 anglers also provided lengths of tagged fish through a reward scheme. We assessed body 128 condition from wild and stocked rainbow trout angled at fishing competitions between 2004 129 and 2010. We measured mass to the nearest 10 g and fork length to the nearest 5 mm. Use of 130 angler returns likely introduced biases into our data that cannot be fully quantified without 131 independently collected data. We assume that biases were similar among lakes. 132 133

Growth and condition of rainbow trout 134 To model trout growth, we fitted the von Bertalanffy growth equation to length-at-age data 135 for rainbow trout in different lakes and seasons using least-squares regression: 136

]1[ )( 0ttKt eLL −−

∞ −= (1) 137

where t is age (yr), Lt is fork length at age t (mm), and L∞ (mm), K (mm yr-1), and t0 138 (yr) are parameters. We used the mean or modal length at release, depending on available 139 data, as the initial data point, and weighted values based on the actual number of measured 140 fish. We defined the day that fish were released into the lake as t = 0. To compare the growth 141 rates of fish in various lakes, we calculated the instantaneous growth rate at 480 mm (the 142

approximate fork length of fish at age 2) using the Gulland and Holt (1959) presentation of 143 von Bertalanffy’s growth equation: 144

)(/ tLLKdtdL −= ∞ (2) 145

where dL/dt is fish growth rate, Lt is fork length at time t, and K is a coefficient. 146 Because the parameters of the von Bertalanffy equation may be influenced by the absence of 147 older fish or by inadequate sample numbers (Knight 1968), we excluded cohorts from this 148 analysis if the lengths of fewer than 10 fish were returned, if fewer than five fish >3 years old 149 were returned, or if L∞ or K fell outside of three standard deviations from the overall mean. 150 This excluded 40% of cohorts released. Calculating the growth rate at 480 mm allowed us to 151 match growth of a specific cohort with the conditions in the lake during the preceding year, 152 and reduced bias associated with variation of fish length at release. We then used the fitted 153 von Bertalanffy relationships to calculate the length of rainbow trout in each cohort at age 2 154 and 5 years. 155

To compare the body condition of rainbow trout in lakes Okataina, Rotorua, Rotoiti, 156 and Tarawera, we used an ANCOVA of log10-transformed weight with log10-transformed 157 length as a covariate. We chose these lakes because they are the most heavily fished by 158 anglers and represent a range of trophic states. Comparisons of weight-at-length can be used 159 to compare the body condition of fish populations, e.g., among locations, times of the year or 160 following environmental perturbations. Fish of the same length with greater body mass have 161 more energy stored, and are in a healthier physiological state. 162 163

Relationships between growth rates and lake characteristics 164 To determine the factors that were important in controlling trout growth rates in different 165 lakes, we assessed relationships between growth rates at 480 mm and L∞ of rainbow trout and 166 multiple environmental factors. We measured temperature, DO, and specific conductivity 167 monthly at a mid-lake station in each lake, at 1 or 2 m intervals throughout the water column, 168 with a conductivity-temperature-depth (CTD) profiler (SBE 19 plus, Seabird Electronics) 169 with CTD-mounted DO sensors (Seabird Electronics). We averaged measurements from each 170 July to the following June, to provide information on average annual conditions, to 171 incorporate the period of summer (December-February) stratification, and to coincide with 172 the first-year growth of each trout cohort in the lakes. 173

We calculated the mid-metalimnion depth (hereafter referred to as the thermocline 174 depth) using temperature profiles taken with the aforementioned CTD profiler during summer 175 stratification (December-February), by finding the point at which dT/dz was maximum, 176 where T = temperature and z = lake depth. We defined the volumes of the epilimnion and 177 hypolimnion during summer stratification as the volumes of water above and below the 178 thermocline (December-February), and calculated these using bathymetry data. Samples for 179 total nitrogen (TN), total phosphorus (TP), and chl a were taken monthly, coinciding with 180 CTD profiles (Scholes 2009, 2010). In lakes that stratified (Okareka, Okataina, Rotoiti, 181 Rotoma, Rotorua, and Tarawera), we collected water samples with an integrated tube sampler 182 that sampled water from the surface to a depth 1 m above the thermocline. We collected one 183 sample from directly below the thermocline and one from 1 m above the bottom sediment 184 with a Schindler-Patalas trap. We used an average of all three samples in our analyses. In 185 lakes Rerewhakaaitu and Rotoehu , which are frequently mixed, we took integrated tube 186 samples from the surface to 1 m above the bottom sediment. We determined concentrations 187 of TN, TP and chl a using standard methods (Eaton and Franson 2005). We measured water 188 transparency with a 20 cm diameter Secchi disk. Because catchment land use affects 189 productivity and water quality of lakes, we also factored catchment land use type into the 190

analysis by standardising the area of each catchment land use (forest, agriculture, and urban; 191 Scholes 2009) by dividing by the lake surface area. 192

To characterise monthly changes in suitable thermal habitat volume, we calculated the 193 volume of water in each lake that was <21ºC with monthly temperature profiles and lake 194 bathymetry data. We then documented volume during the month when suitable habitat 195 volume was at its minimum in each lake (i.e., the “worst case scenario” for trout). We used 196 this temperature threshold because rainbow trout have been observed to avoid temperatures 197 above 21ºC in lakes (Overholtz et al. 1977, Stables and Thomas 1992), and reduced growth 198 rates were observed in rainbow trout at 22ºC and 25ºC compared to 19ºC (Myrick and Cech 199 2000). We calculated the minimum habitat available with DO concentrations above 6 mg L-1 200 in a similar way; we chose this concentration because salmonids begin to exhibit symptoms 201 of oxygen stress below 6 mg L-1 DO (Davis 1975). We also calculated a composite measure 202 of the habitat volume with both suitable DO and temperature. We calculated near-bottom 203 oxygen concentration from DO measurements in the bottom 5 m of water during 204 stratification, and in the bottom 2 m of the shallowest lakes (Rerewhakaaitu and Rotoehu). 205

Population size may also influence growth of fish, but comparing density-dependent 206 effects among lakes is difficult because differences in environmental factors among lakes 207 may confound comparisons. Hence, density-dependent effects on trout growth in Lake 208 Rotoiti were the subject of a concurrent bioenergetics study (Blair et al. 2013). Nevertheless, 209 we incorporated the following estimates of trout density in our analyses: (1) the number of 210 trout released from the hatchery per year in each particular lake, (2) the number of trout 211 released per year per hectare of lake surface area, and (3) the length of trout upon release. We 212 also calculated the ratio of stocked to wild trout in angler catches, using records gained from 213 creel surveys every summer over the study period (Table 2). This provides only an 214 approximate index of the relative abundance of wild trout. Little is known about the 215 population abundance of wild trout or their susceptibility to fishing mortality by angling; 216 differences in mortality rates between wild and stocked trout or among years may affect this 217 index. 218

219

Data analysis 220 We compared growth rates of trout released in autumn and spring in individual lakes with 221 paired Student’s t-tests. We assessed the differences in L∞ and growth rate at 480 mm 222 between deep and shallow lakes using a Mann-Whitney U test. We used Spearman’s rank 223 correlations to examine the relationships between trout growth rate, L∞, and environmental 224 factors in different lakes. The mean of all yearly measurements in each lake was treated as an 225 independent data point for this analysis. Lake Tutira was not included in all the correlation 226 analyses due to insufficient environmental data. Body condition of rainbow trout in lakes 227 Rotoiti, Rotorua, Okataina and Tarawera was compared using an ANCOVA, with log10-228 transformed weight data as the dependent variable, lake as the categorical factor and log10-229 transformed length as the covariate. 230

To predict the conditions that would produce higher trout growth rate and L∞ in our 231 study lakes, and to determine whether inter-annual variability within lakes influenced growth, 232 we used classification and regression trees (CART) to classify data from different lakes and 233 years. Classification and regression tree analysis is a non-parametric statistical method that 234 has advantages over parametric regression because it does not assume that data are 235 homogeneous or homoscedastic, or that relationships between predictor and response 236 variables are linear. Further, CART can deal with missing values, interactions among 237 variables, and mixtures of continuous and categorical data, and can uncover relationships that 238 may be missed by linear models (Breiman et al. 1984, De’ath and Fabricius 2000). We built a 239

regression tree using the CART function in Statistica, which uses the procedure of Breiman et 240 al. (1984). The V-fold cross-validation procedure found the optimal tree using the V-fold 241 cross-validation procedure; this tests trees of multiple sizes by applying them to random 242 subsets of the data, finding the tree with the highest average accuracy for cross-validated 243 predicted classifications. 244

For the CART analysis, we included variables in the model only if (1) they were 245 measured directly and were not derived from another, and (2) they directly reflected 246 conditions in the lake (i.e., catchment area and land use variables were excluded). We showed 247 the importance of each variable for predicting trout growth as a relative percentage. 248 STATISTICA calculates relative importance by summing using the resubstitution estimate 249 (the proportion of cases that are misclassified) for all nodes in the tree, and expressing the 250 sums as a percentage of the largest sum (i.e., that of the most important variable). Up to four 251 surrogate variables indicated alternative decision rules for each node, and we calculated the 252 percentage agreement with the chosen variable in the allocation of cases. Surrogate variables 253 are used in the case of missing data points; our dataset was missing 1 year of conductivity 254 measurements for Lake Rotoiti but was otherwise complete. 255

256

Results 257

The maximum age of returned fish was, on average, 4.36 yr for fish released in spring, and 258 4.17 yr for fish released in autumn (Table S1 in Supplementary Information). Slightly more 259 fish were caught in summer or autumn in most lakes. 260

Our results indicate clear effects of lake trophic state and depth on trout growth rates. 261 Rainbow trout in deeper lakes showed higher values of L∞ than those in shallow lakes (Table 262 3, Mann-Whitney (M-W) U test: Z = -4.56, P < 0.001, n (deep) = 6, n (shallow) = 3). Growth 263 rates of rainbow trout at 480 mm were also higher in deep lakes than in shallow lakes (M-W 264 U test: Z = -6.59, P < 0.001, n (deep) = 6, n (shallow) = 3). Growth rate at 480 mm was 265 positively correlated with L∞ (Spearman rank rs = 0.909, P < 0.001, n = 9), as well as with 266 lengths at age 2 and 5 years (rs = 0.917 and 0.983 respectively, n = 9, P < 0.001). The season 267 in which trout yearlings were released into the lakes (spring or autumn) did not significantly 268 affect subsequent mean growth rates in lakes Okataina, Rotoiti, Tarawera or Tutira (Student’s 269 paired t-tests, P > 0.05, n = 8). 270

Body condition of rainbow trout, assessed by an ANCOVA of weight with length as 271 the covariate, differed significantly among all lakes. Condition was highest in mesotrophic 272 Lake Rotoiti and lowest in oligotrophic Lake Tarawera (Table S2; Tukey’s HSD). 273

Among lake morphometry variables, growth rate showed significant relationships 274 only with mean and maximum depth (Fig. 1a-c). However, the direction of nearly all 275 correlations between trout growth variables and variables related to lake morphometry 276 suggest that trout growth rates increase with increasing lake size (i.e., area, volume, and 277 depth) (Fig. 1). Increased trout growth rates were associated with larger hypolimnetic 278 volumes, lower epilimnion:hypolimnion ratios, and greater summer thermocline depths (Fig. 279 2a-c). Growth rates declined significantly with increasing TN, but did not show a significant 280 correlation with either temperature, TP, stocking rate, or conductivity (Fig. 2e-i). Trout 281 growth rates were positively correlated with an indicator of habitat volume: the summer 282 minimum volume of the lake with suitable temperature and DO (Fig. 2d). In contrast, trout 283 growth rate was not significantly correlated with averaged temperature and DO 284 measurements throughout the water column (Table S3). Relationships between L∞ and tested 285 variables were similar to the relationships with growth rate (Table S3). 286

The CART analysis showed that variables associated with lake volume and the 287 volume of suitable habitat were the most important predictors of trout growth rates in 288

different lakes and years (Table 4). Stocking rate and mean lake temperature were also 289 important variables, whereas trophic state indicators such as chl a and TP were less 290 important. Consequently, the optimal tree design selected hypolimnetic volume as the 291 variable most effectively separating low from high trout growth rates (Fig. 3). Within these 292 two groups, indicators of trophic state were important for further partitioning of trout growth 293 rates, but in different ways: in lakes with high hypolimnetic volume, higher chl a 294 concentrations were associated with higher growth rates, whereas in lakes with low 295 hypolimnetic volume, high conductivity was associated with lower growth rates. Also, 296 although conductivity was correlated with growth rate in low-volume lakes, there was no 297 such correlation among all lakes in the data set (Table S3). As possible alternatives for the 298 variables in the tree, the model also selected several surrogate variables (Table 5). The 299 surrogates for hypolimnion volume all relate to volume, depth, DO or temperature and hence 300 indicate habitat quality or volume, whereas the surrogates for chl a are all indicators of 301 trophic state. However, temperature was chosen as the variable that was most closely related 302 with conductivity, and other surrogate variables for this node had little association with 303 conductivity. 304 305

Discussion 306

Relationships between growth rates and lake characteristics 307 Trout growth parameters increased along a lake productivity gradient in deep lakes of low to 308 medium productivity, but declined with increasing productivity in shallow, highly productive 309 lakes. Changes in body condition of rainbow trout with increasing lake productivity in deep 310 lakes were consistent with patterns in growth rate. Increases in nutrients may be beneficial for 311 growth of rainbow trout when sufficient habitat is present (i.e., in deep lakes), but severe 312 eutrophication may be detrimental for trout growth where habitat is limited by high 313 temperature and low DO (i.e., in shallow lakes). This result expands upon observations by 314 Budy et al. (2011), who found that the growth rate potential of coolwater species (walleye 315 Sander vitreus and crappie Pomoxis nigromaculatus) decreased with increasing chl a 316 concentrations in a warm lake. 317

Habitat availability is a key factor structuring fish populations; for example, 318 temperature and DO concentrations limit habitat use by salmonids in lakes (Swales 2006, 319 Plumb and Blanchfield 2009), and the availability of suitable thermal habitat in lakes 320 influences fish growth rate potential (Mason et al. 1995, Walline et al. 2000, Budy et al. 321 2011) and total sustained yield (Christie and Regier 1988). Our study expands upon this 322 previous work by showing that reductions in habitat during summer, measured by the 323 minimum volume of water with suitable DO and temperature, affect growth rates of rainbow 324 trout in lakes. In the shallow lakes in our study, near-bottom waters can exceed 20ºC during 325 summer (Burger et al. 2007, P. Scholes, unpubl. data); creating unsuitable temperature 326 conditions for rainbow trout throughout the water column (Overholtz et al. 1977, Stables and 327 Thomas 1992). By providing a refuge, cold tributaries may mitigate the effects of warm 328 temperatures in shallow Lake Rotorua: Hamurana Stream (mean flow 256 000 m3 d-1) and 329 Awahou Stream (256 000 m3 d-1) provide inputs of water below 12ºC (Rutherford et al. 1996, 330 Boubée et al. 2006). Thermal refuge habitats may offer the best conditions for growth for 331 coldwater species in warm lakes during summer (Budy et al. 2011). However, limited volume 332 of suitable habitat may still have substantial effects on growth rates; these effects are not well 333 documented yet and require further detailed investigations. 334

In shallow, eutrophic lakes such as Rotorua and Rotoehu, periodic stratification 335 events cause intermittent hypolimnetic deoxygenation (Burger et al. 2007, Trolle et al. 2011), 336 which probably limits rainbow trout habitat further, especially if high temperatures and low 337

DO coincide. If fish are forced into areas with suboptimal temperature conditions and 338 minimal prey concentrations during thermal stratification, growth rate potential may be 339 reduced (Roberts et al. 2009, Arend et al. 2011). This effect is likely to be most pronounced 340 in lakes that lack significant cold input streams, such as lakes Rerewhakaaitu and Rotoehu. 341 Areas of low DO and high temperature in Lake Rotoehu coincided on several occasions 342 during summer (data not shown), which may account for the low trout growth rates in this 343 lake. 344

Deoxygenation of the near-bottom waters of lakes may not be harmful for trout 345 populations, provided there is sufficient suitable habitat outside the deoxygenated zone. Trout 346 growth rates were negatively correlated with mean water column and near-bottom DO 347 concentrations in our study. This seems to be paradoxical, given the positive correlations 348 observed between trout growth rates and habitat measures; however, other studies show that 349 hypolimnetic deoxygenation in deep lakes may increase fish growth when conditions in the 350 upper water column are favourable, by concentrating prey. For example, hypoxia increased 351 growth potential of Lake Erie walleye (Sander vitreus) by concentrating prey in areas with 352 suitable light, DO and temperature conditions (Brandt et al. 2011). This may be the case for 353 the deeper Rotorua lakes, as the main prey of rainbow trout in New Zealand lakes, the 354 common smelt (Retropinna retropinna; Rowe 1984), is highly sensitive to reduced DO 355 concentrations and is therefore likely to avoid areas of low DO (Dean and Richardson 1999). 356 Further investigations into the effects of temperature on the distribution of rainbow trout and 357 their prey (e.g., using hydroacoustics) are needed to confirm the link between habitat and 358 growth rates. 359

Habitat quality is linked to population density, and the negative correlation between 360 trout stocking rate and growth rate indicates possible density-dependent effects on trout 361 growth in our study system. The spread of data evident in this relationship (especially at low 362 stocking densities) indicates the presence of confounding factors, such as environmental 363 conditions and the population size of wild trout. Reliable assessment of density dependent 364 effects requires estimates of the entire trout population, which are currently lacking in our 365 study lakes. However, reduced growth rates in Lake Okataina between 2004 and 2007, as 366 seen in our regression tree, may have been due to a change in angling size regulations: during 367 this time, the minimum catch size was raised from 35 to 58 cm, meaning that fewer trout 368 were harvested, which probably led to higher than normal population densities and density-369 dependent effects on growth. 370 371

Responses of rainbow trout growth rates 372 Persistence of populations in changing environments (e.g., lakes that are becoming warmer) 373 depends on their ability to respond. Responses to environmental change may involve 374 distribution changes, phenotypic plasticity, or evolution, and determining the mechanism 375 responsible for observed adaptation is important for predicting the effects of perturbations 376 such as climate change (Gienapp et al. 2007). Phenotypic plasticity can allow a population to 377 adapt to environmental changes in the short term (Berteaux et al. 2004). We observed wide 378 variation in growth rates, L∞, and body condition of rainbow trout that originated from the 379 same genetic source and were raised under identical hatchery conditions, providing evidence 380 for phenotypic plasticity of growth rates and final size. Similarly, Reed et al. (2010) found 381 that genetically similar, hatchery-raised sockeye salmon (Oncorhynchus nerka) grew larger 382 when stocked into a warm, productive lake compared to a cooler, less productive lake. Future 383 studies should clarify the effects of lake eutrophication and warm temperatures on other 384 phenotypically plastic life history traits, such as age at maturity and timing of spawning 385

(Crozier et al. 2008), to help predict the longer term consequences of global environmental 386 change. 387 388

Climate change and catchment management implications 389 By directly examining growth of coldwater piscivores in near-stress situations, e.g., warm-390 temperate lakes such as those in New Zealand’s North Island, we can infer the likely impacts 391 of future temperature increases on coldwater piscivores in cooler areas. In most temperate 392 lakes, climate change is predicted to cause increases in water temperature (Gerten and Adrian 393 2002). Coldwater species living in shallow lakes are likely to experience greater reductions in 394 favourable habitat than in deep lakes (Stefan et al. 2001, Fang et al. 2004), and this is 395 expected to negatively affect their growth (this study) and abundance (Jeppesen et al. 2012). 396 Though climate change in New Zealand is predicted to be less severe than in many other 397 locations, increases in the duration of lake stratification are likely to cause increased 398 deoxygenation of near-bottom waters, which may affect the availability of fish habitat 399 (Hamilton et al. 2012). Our results suggest that climate warming is likely to constrain growth 400 of coldwater piscivores in shallow lakes, particularly those that lack thermal refugia. 401 Furthermore, the effects of eutrophication in lakes are predicted to intensify with climate 402 change (Jeppesen et al. 2009, Trolle et al. 2011), and our results show that slowing or 403 reversing the effects of cultural eutrophication in shallow, warm-temperate lakes is important 404 for conserving their rainbow trout fisheries. In situations where managers can prioritise water 405 bodies for trout stocking, deeper lakes provide the best opportunity for conserving these 406 economically and culturally important fisheries. Providing that further eutrophication of these 407 lakes is prevented, trout should have sufficient habitat around and below the thermocline that 408 is well-oxygenated and within a suitable temperature range. 409

410

Acknowledgements 411

The authors thank M. Osborne, A. Uytendaal, and staff at Fish & Game for collecting data, 412 D. Özkundakci for advice on limnological data analysis, and D. Hamilton, N. Ling, M. 413 Pingram, J. Abell and anonymous reviewers for helpful comments. Funding was provided by 414 Bay of Plenty Regional Council, Lake Ecosystem Restoration New Zealand Research 415 Contract UOW X0505 from the New Zealand Ministry of Science and Innovation, and the 416 University of Waikato. 417 418

References 419

Arend, K. K., Beletsky, D., DePinto, J. V., Ludsin, S. A., Roberts, J. J., Rucinski, D. K., 420 Scavia, D., Schwab, D. J., and Höök, T. O. 2011. Seasonal and interannual effects of 421 hypoxia on fish habitat quality in central Lake Erie. Freshwater Biology 56: 366–383. 422

Beauchamp, D. A., Baldwin, C. M., Vogel, J. L., and Gubala, C. P. 1999. Estimating diel, 423 depth-specific foraging opportunities with a visual encounter rate model for pelagic 424 piscivores Canadian Journal of Fisheries and Aquatic Sciences 56: 128–139. 425

Berteaux, D., Réale, D., McAdam, A.G., and Boutin, S. 2004. Keeping pace with fast climate 426 change: Can arctic life count on evolution? Integrative and Comparative Biology 44: 427 140–151. 428

Blair, J.M., Hicks, B.J., Pitkethley, R.J., Ling, N., Ostrovsky, I., and Rowe, D.K. 2013. A 429 bioenergetic assessment of the influence of stocking practices on rainbow trout 430 (Oncorhynchus mykiss) growth and consumption in a New Zealand lake. Freshwater 431 Biology. doi: 10.1111/fwb.112100. 432

Boisclair, D., and Rasmussen, J.B. 1996. Empirical analysis of the influence of 433 environmental variables associated with lake eutrophication on perch growth, 434 consumption, and activity rates. Annales Zoologici Fennici 33: 507–515. 435

Boubée, J., Bowman, E., and Rowe, D. K. 2006. Effects of the cold, spring-fed inlet streams 436 on rainbow trout distribution in Lake Rotorua. NIWA Client Report HAM 2006-121, 437 National Institute of Water and Atmospheric Research, Hamilton. Available from 438 http://old.boprc.govt.nz/TechReports/TechReports-061001-EffectStreamsonTrout.pdf 439 [Accessed 13 February 2012]. 440

Brandt, S. B., Costantini, M., Kolesar, S., Ludsin, S. A., Mason, D. M., Rae, C. M., and 441 Zhang, H. 2011. Does hypoxia reduce habitat quality for Lake Erie walleye (Sander 442 vitreus)? A bioenergetics perspective. Canadian Journal of Fisheries and Aquatic 443 Sciences 68: 857–879. 444

Breiman, L., Friedman, J.H., Olshen, R.A., and Stone, C.G. 1984. Classification and 445 regression trees. Wadsworth International Group, Belmont, CA. 446

Budy, P., Thiede, G.P., Luecke, C., and Schneidervin, R.W. 2009. Warmwater and coldwater 447 fish in two-story standing waters. In Standard methods for sampling North American 448 freshwater fishes. Edited by S.A. Bonar, W.A. Hubert and D.W. Willis. American 449 Fisheries Society, Bethesda, MD. 450

Budy, P., Baker, M., and Dahle, S. 2011. Predicting fish growth potential and identifying 451 water quality constraints: a spatially-explicit bioenergetics approach. Environmental 452 Management 48: 691–709. 453

Burger, D. F., Hamilton, D. P., Pilditch, C. A., and Gibbs, M. M. 2007. Benthic nutrient 454 fluxes in a eutrophic, polymictic lake. Hydrobiologia 584: 13–25. 455

Burns, N.M., Rutherford, J.C., and Clayton, J.S. 1999. A monitoring and classification 456 system for New Zealand lakes and reservoirs. Lake and Reservoir Management 15: 457 255–271. 458

Christie, G.C., and Regier, H.A. 1988. Measures of optimal thermal habitat and their 459 relationship to yields for four commercial fish species. Canadian Journal of Fisheries 460 and Aquatic Sciences 45: 301–314. 461

Coutant, C. C. 1985. Striped bass, temperature, and dissolved-oxygen – a speculative 462 hypothesis for environmental risk. Transactions of the American Fisheries Society 463 114: 31–61. 464

Crozier, L.G., Hendry, A.P., Lawson, P.W., Quinn, T.P., Mantua, N.J., Battin, J., Shaw, R.G., 465 and Huey, R.B. 2008. Potential responses to climate change in organisms with 466 complex life histories: evolution and plasticity in Pacific salmon. Evolutionary 467 Applications 1: 252–270. 468

Davis, J. C. 1975. Minimal dissolved oxygen requirements of aquatic life with emphasis on 469 Canadian species: a review. Journal of the Fisheries Research Board of Canada 32: 470 2295–2332. 471

Dean, T. L., and Richardson, J. 1999. Responses of seven species of native freshwater fish 472 and a shrimp to low levels of dissolved oxygen. New Zealand Journal of Marine and 473 Freshwater Research 33: 99–106. 474

De'ath, G., and Fabricius, K.E. 2000. Classification and regression trees: A powerful yet 475 simple technique for ecological data analysis. Ecology 81: 3178–3192. 476

Downing, J. A., and Plante, C. 1993. Production of fish populations in lakes. Canadian 477 Journal of Fisheries and Aquatic Sciences 50: 110–120. 478

Eaton, A. D. and Franson, M. A. H. 2005. Standard methods for the examination of water and 479 wastewater. American Public Health Association, Washington D.C. 480

Fang, X., Stefan, H.G., Eaton, J.G., McCormick, J.H., and Alam, S.R. 2004. Simulation of 481 thermal/dissolved oxygen habitat for fishes in lakes under different climate scenarios 482 Part 1. Cool-water fish in the contiguous US. Ecological Modelling 172: 13–37. 483

Fee, E. J. 1979. A relation between lake morphometry and primary productivity and its use in 484 interpreting whole-lake eutrophication experiments. Limnology and Oceanography 485 24: 401–416. 486

Genkai-Kato, M., and Carpenter, S. R. 2005. Eutrophication due to phosphorus recycling in 487 relation to lake morphometry, temperature, and macrophytes. Ecology 86: 210–219. 488

Gerten, D., and Adrian, R. 2002. Effects of climate warming, North Atlantic Oscillation, and 489 El Niño-Southern Oscillation on thermal conditions and plankton dynamics in 490 northern hemispheric lakes. TheScientificWorldJOURNAL 2: 586–606. 491

Gienapp, P., Teplitsky, C., Alho, J.S., Mills, J.A., and Merilä, J. 2008. Climate change and 492 evolution: disentangling environmental and genetic responses. Molecular Ecology 17: 493 167–178. 494

Gulland, J. A., and Holt, S. J. 1959. Estimation of growth parameters for data at unequal time 495 intervals. Journal du Conseil international pour l'exploration de la mer. 25: 47–49. 496

Haddix, T., and Budy, P. 2005. Factors that limit growth and abundance of rainbow trout 497 across ecologically distinct areas of Flaming Gorge Reservoir, Utah-Wyoming. North 498 American Journal of Fisheries Management 25: 1082–1094. 499

Håkanson, L. 2005. The importance of lake morphometry for the structure and function of 500 lakes. Int. Rev. Hydrobiologia 90: 433–461. 501

Hamilton, D.P., McBride, C., Ozkundakci, D., Schallenberg, M., Verburg, P., de Winton, M., 502 Kelly, D., Hendy, C., and Ye, W. 2012. Effects of climate change on New Zealand 503 lakes. In Climate Change and Inland Waters: Impacts and Mitigation Approaches for 504 Ecosystems and Society. Edited by C.R. Goldman, M. Kumagai and R.D. Robarts. 505 Wiley-Blackwell. 506

Hyatt, K. D., Mcqueen, D. J., Shortreed, K. S., and Rankin, D. P. 2004. Sockeye salmon 507 (Oncorhynchus nerka) nursery lake fertilization: review and summary of results. 508 Environmental Reviews 12: 133–162. 509

Jeppesen, E., Kronvang, B., Meerhoff, M., Søndergaard, M., Hansen, K. M., Andersen, H. E., 510 Lauridsen, T. L., Liboriussen, L., Beklioglu, M., Özen, A., and Olesen, J. E. 2009. 511 Climate change effects on runoff, catchment phosphorus loading and lake ecological 512 state, and potential adaptations. Journal of Environmental Quality 38: 1930–1941. 513

Jeppesen, E., Mehner, T., Winfield, I., Kangur, K., Sarvala, J., Gerdeaux, D., Rask, M., 514 Malmquist, H., Holmgren, K., Volta, P., Romo, S., Eckmann, R., Sandström, A., 515 Blanco, S., Kangur, A., Ragnarsson Stabo, H., Tarvainen, M., Ventelä, A.-M., 516 Søndergaard, M., Lauridsen, T., and Meerhoff, M. 2012. Impacts of climate warming 517 on the long-term dynamics of key fish species in 24 European lakes. Hydrobiologia 518 694: 1–39. 519

Knight, W. 1968. Asymptotic growth: an example of nonsense disguised as mathematics. 520 Journal of the Fisheries Research Board of Canada 25: 1303–1307. 521

Leach, J. H., Dickie, L. M., Shuter, B. J., Borgmann, U., Hyman, J., and Lysack, W. 1987. A 522 review of methods for prediction of potential fish production with application to the 523 Great Lakes and Lake Winnipeg. Canadian Journal of Fisheries and Aquatic Sciences 524 44: 471–485. 525

Mason, D.M., Goyke, A., and Brandt, S.B. 1995. A spatially explicit bioenergetics measure 526 of habitat quality for adult salmonines: comparison between Lakes Michigan and 527 Ontario. Canadian Journal of Fisheries and Aquatic Sciences 52: 1572–1583. 528

Myrick, C. A., and Cech, J. J. 2000. Temperature influences on California rainbow trout 529 physiological performance. Fish Physiology and Biochemistry 22: 245–254. 530

Oglesby, R. T. 1977. Relationships of fish yield to lake phytoplankton standing crop, 531 production and morphoedaphic factors. Journal of the Fisheries Research Board of 532 Canada 34: 2271–2279. 533

Overholtz, W. J., Fast, A. W., Tubb, R. A., and Miller, R. 1977. Hypolimnion oxygenation 534 and its effects on the depth distribution of rainbow trout (Salmo gairdneri) and 535 gizzard shad (Dorosoma cepedianum). Transactions of the American Fisheries 536 Society 106: 371–375. 537

Pankhurst, N. W., and King, H. R. 2010. Temperature and salmonid reproduction: 538 implications for aquaculture. Journal of Fish Biology 76: 69–85. 539

Plante, C., and Downing, J. 1993. Relationship of salmonine production to lake trophic state 540 and temperature. Canadian Journal of Fisheries and Aquatic Sciences 50: 1324–1328. 541

Plumb, J. M., and Blanchfield, P. J. 2009. Performance of temperature and dissolved oxygen 542 criteria to predict habitat use by lake trout (Salvelinus namaycush). Canadian Journal 543 of Fisheries and Aquatic Sciences 66: 2011–2023. 544

Ranåker, L., Jönsson, M., Nilsson, P.A., and Brönmark, C. 2012. Effects of brown and turbid 545 water on piscivore–prey fish interactions along a visibility gradient. Freshwater 546 Biology 57: 1761–1768. 547

Reed, T. E., Martinek, G., and Quinn, T. P. 2010. Lake-specific variation in growth, 548 migration timing and survival of juvenile sockeye salmon Oncorhynchus nerka: 549 separating environmental from genetic influences. Journal of Fish Biology 77: 692–550 705. 551

Roberts, J. J., Höök, T. O., Ludsin, S. A., Pothoven, S. A., Vanderploeg, H. A., and Brandt, 552 S. B. 2009. Effects of hypolimnetic hypoxia on foraging and distributions of Lake 553 Erie yellow perch. Journal of Experimental Marine Biology and Ecology 381, 554 Supplement: S132–S142. 555

Robinson, J. M., Josephson, D. C., Weidel, B. C., and Kraft, C. E. 2010. Influence of variable 556 interannual summer water temperatures on brook trout growth, consumption, 557 reproduction, and mortality in an unstratified Adirondack lake. Transactions of the 558 American Fisheries Society 139: 685–699. 559

Rowe, D.K. 1984. Factors affecting the foods and feeding patterns of lake-dwelling rainbow 560 trout (Salmo gairdnerii) in the North Island of New Zealand. New Zealand Journal of 561 Marine and Freshwater Research 18: 129–141. 562

Rutherford, J. C., Dumnov, S. M., and Ross, A. H. 1996. Predictions of phosphorus in Lake 563 Rotorua following load reductions. New Zealand Journal of Marine and Freshwater 564 Research 30: 383–396. 565

Scholes, P. 2009. Rotorua Lakes Water Quality Report 2009. Environmental Publication 566 2009/12, Bay of Plenty Regional Council, Whakatane. Available from 567 http://www.boprc.govt.nz/media/33520/Report-091221-568 RotoruaLakesWaterQualityReport2009.pdf [Accessed 30 July 2012]. 569

Scholes, P. 2010. 2009/2010 Rotorua Lakes TLI Update. Environmental Publication 2010/18, 570 Bay of Plenty Regional Council, Whakatane. Available from 571 http://www.boprc.govt.nz/media/140219/environmentalpublication2011_17201011rot572 orualakestliupdate_final.pdf [Accessed 30 July 2012]. 573

Schindler, D. W. 2001. The cumulative effects of climate warming and other human stresses 574 on Canadian freshwaters in the new millennium. Canadian Journal of Fisheries and 575 Aquatic Sciences 58: 18–29. 576

Stables, T. B., and Thomas, G. L. 1992. Acoustic measurement of trout distributions in Spada 577 Lake, Washington, using stationary transducers. Journal of Fish Biology 40: 191–203. 578

Stefan, H. G., Fang, X., and Eaton, J. G. 2001. Simulated fish habitat changes in North 579 American lakes in response to projected climate warming. Transactions of the 580 American Fisheries Society 130: 459–477. 581

Swales, S. 2006. A review of factors affecting the distribution and abundance of rainbow 582 trout (Oncorhynchus mykiss Walbaum) in lake and reservoir systems. Lake and 583 Reservoir Management 22: 167–178. 584

Trolle, D., Hamilton, D. P., Pilditch, C. A., Duggan, I. C., and Jeppesen, E. 2011. Predicting 585 the effects of climate change on trophic state of three morphologically varying lakes: 586 implications for lake restoration and management. Environmental Modelling and 587 Software. 26: 354–370. 588

Walline, P. D., Tyler, J. A., Brandt, S. B., Ostrovsky, I., and Jech, J. M. 2000. Lavnun 589 abundance: how changes may affect consumption of Lake Kinneret zooplankton. 590 Fundamental and Applied Limnology 55: 493–511 591

Wells, B. K., Grimes, C. B., Sneva, J. G., Mcpherson, S., and Waldvogel, J. B. 2008. 592 Relationships between oceanic conditions and growth of Chinook salmon 593 (Oncorhynchus tshawytscha) from California, Washington, and Alaska, USA. 594 Fisheries Oceanography 17: 101–125. 595

Tables Table 1. Morphometry and limnological characteristics of the nine study lakes in the North Island of New Zealand. Values are means for 2002–2010 ± SE, except for Lake Tutira, where data are from 2008–2010.

Lake

Mean depth (m)

Area (km2)

Mixing regime

Trophic state

Mean temperature

(ºC)

Annual temperature range (ºC)

Chl a (mg m-3)

Secchi depth (m)

TP (mg P m-3)

TN (mg N m-3)

Rere. 6 5.9 P Me 14.3 ± 0.2 8.1–21.6 3.7 ± 0.2 5.8 ± 0.2 9.7 ± 0.5 399.5 ± 7.9 Rotoehu 8 8.1 P Eu 15.3 ± 0.1 9.2–22.6 10.8 ± 0.8 2.9 ± 0.1 39.8 ± 2.0 376.5 ± 9.6 Rotorua 11 80.8 P Eu 15.0 ± 0.1 9.3–21.7 21.3 ± 0.9 2.4 ± 0.1 37.9 ± 0.8 458.4 ± 8.4 Okareka 20 3.3 M Me 12.4 ± 0.2 9.3–22.2 4.1 ± 0.2 7.7 ± 0.2 8.4 ± 0.3 212.6 ± 3.8 Tutira 21 1.7 M Me 12.1 ± 0.1 9.0–22.7 4.4 ± 1.1 3.9 ± 0.3 42.2 ± 17.4 493.9 ± 13.7 Rotoiti 31 34.6 M Me 12.8 ± 0.1 10.8–22.1 12.2 ± 0.9 4.4 ± 0.1 40.4 ± 0.7 356.6 ± 4.5 Rotoma 37 11.2 M Ol 12.5 ± 0.3 10.6–21.3 1.3 ± 0.1 12.8 ± 0.3 4.6 ± 0.2 141.2 ± 4.0 Okataina 39 10.8 M Ol 12.5 ± 0.3 10.7-21.4 2.3 ± 0.1 10.5 ± 0.3 11.5 ± 0.5 135.0 ± 4.2 Tarawera 50 41.7 M Ol 12.5 ± 0.2 10.9–21.4 1.5 ± 0.1 8.7 ± 0.2 13.9 ± 0.6 114.0 ± 4.0

Note: The temperature range is the mean of the annual temperature minima-the mean of the annual temperature maxima. Rere., Rerewhakaaitu; P, polymictic; M, monomictic; Ol, oligotrophic; Me, mesotrophic; Eu, eutrophic. Sources: Lake Tutira; A. Uytendaal, Hawke’s Bay Regional Council, Napier, pers. comm., 2011, all other lakes; Scholes 2009, 2010. Trophic state based on the New Zealand Trophic Level Index for lakes (Burns et al. 1999)

Table 2. Stocking rates, percentage of stocked rainbow trout in recreational catch, and fork length of trout at release in lakes in the North Island, New Zealand.

Lake

Stocking rate (fish year-1)

Stocking rate (fish ha-

1 year-1)

Percentage of stocked trout in

catch Length of trout stocked

(mm) Rerewhakaaitu 7375 ± 100 1272 ± 19 68.6 ± 5.4 Spring: 190 ± 7 Rotoehu 4380 ± 388 596 ± 54 Spring: 190 ± 7 Rotorua 1000 ± 0 12 ± 0 2.0 ± 0.6 Spring: 190 ± 7 Okareka 5000 ± 683 1515 ± 231 85.0 ± 2.7 Spring: 190 ± 7 Tutira 5346 ± 485 3168 ± 369 Spring: 200 ± 7

Autumn: 160 ± 6 Rotoiti 23812 ± 1114 688 ± 36 60.3 ± 2.6 Spring: 190 ± 8

Autumn: 170 ± 6 Rotoma 5006 ± 369 447 ± 37 64.4 ± 6.1 Spring: 190 ± 7 Okataina 3750 ± 250 347 ± 31 58.0 ± 3.3 Spring: 190 ± 8

Autumn: 160 ± 5 Tarawera 17875 ± 598 429 ± 15 69.7 ± 2.9 Spring: 200 ± 4

Autumn: 160 ± 7 Note: Values are means for 2002–2010 ± SE.

Table 3. Parameters of von Bertalanffy’s growth function (Eqn. 2), growth rates at 480 mm (the length at approximately 2 years), and the mean length at 2 and 5 years, for stocked rainbow trout in deep and shallow lakes in the North Island, New Zealand.

Lake L∞ (mm) K (year-1) t0 (year) Growth rate at 480 mm (mm yr-1)

Length at 2 years (mm)

Length at 5 years (mm)

Shallow lakes (≤11 m mean depth) Rotoehu 508.7 ± 6.8 1.761 ± 0.146 -0.282 ± 0.020 46.3 ± 11.0 451 ± 20.14 508.1 ± 6.7 Rotorua 534.0 ± 7.3 1.579 ± 0.133 -0.304 ± 0.028 94.4 ± 11.8 473.1 ± 14.59 539 ± 10.4 Rerewhakaaitu 558.0 ± 10.0 1.289 ± 0.105 -0.345 ± 0.031 95.1 ± 10.0 454.1 ± 15.21 555.1 ± 9.5 Mean (shallow lakes) 536.7 ± 7.1 1.481 ± 0.095 -0.274 ± 0.015 79.5 ± 7.9 459.1 ± 18.59 535.2 ± 6.8

Deep lakes (≥20 m mean depth)

Tutira 540.6 ± 5.7 1.428 ± 0.107 -0.299 ± 0.023 85 ± 11.0 449.1 ± 34.79 538.2 ± 5.7 Okareka 544.8 ± 11.0 1.519 ± 0.177 -0.311 ± 0.033 85.9 ± 7.0 457.3 ± 23.11 542 ± 9.9 Rotoma 557.1 ± 5.8 1.671 ± 0.142 -0.268 ± 0.022 123.6 ± 5.9 484.9 ± 17.68 556.1 ± 5.5 Tarawera 590.6 ± 3.2 1.515 ± 0.046 -0.232 ± 0.08 164.4 ± 4.1 496.6 ± 16.50 589.4 ± 3.1 Okataina 622.8 ± 9.1 1.278 ± 0.076 -0.262 ± 0.014 175.5 ± 5.4 492.2 ± 18.70 618.6 ± 8.0 Rotoiti 643.1 ± 14.9 1.259 ± 0.109 -0.243 ± 0.026 199.2 ± 3.2 504 ± 9.45 638.9 ± 13.2 Mean (deep lakes) 582.3 ± 7.3 1.568 ± 0.105 -0.302 ± 0.017 140.3 ± 6.1 481.3 ± 29.06 579.8 ± 5.3

Note: Growth parameters represent mean values ± SE for the period from July 2002 to June 2010 (N = 8 years).

Table 4. Importance of variables in predicting trout growth rate in classification and regression tree analysis, expressed as relative percentages.

Variable Relative importance (%) Hypolimnion volume (m3) 100 Mean depth (m) 98 Minimum volume with suitable DO and temp (m3) 86 Lake volume (m3) 72 Minimum volume >6 mg L-1 DO (m3) 72

Stocked fish ha-1 yr-1 70 Mean volume with suitable DO and temp (m3) 69 Mean water temperature (ºC) 66 Min volume <21ºC (m3) 58 Mean DO (mg L-1) 54 Turbidity (NTU) 53 Secchi depth (m) 49 TN (mg N m-3) 41 Conductivity (µS cm-1) 40 Chl a (mg m-3) 37 Size at release (mm) 27 TP (mg P m-3) 24

Table 5. Surrogate values for nodes of regression tree, with the percentage classification agreement of surrogate variables with the variables in the regression tree.

Node Variable

Agreement with chosen variable

(%) Split sign

Split constant

Mean depth (m) 92 – 25.8 Min. vol. suitable DO & temp (m3) 80 – 81663800 Min volume >6 mg L–1 (m3) 64 – 70806575

1. Hypolimnion volume

Mean temperature (°C) 64 + 13.6 2. Conductivity Mean temperature (°C) 71 – 15.1 TP (mg P m–3) 14 – 45.5 Size at release (mm) 14 – 236.7 3. Chl a Mean DO (mg L–1) 80 + 7.87 Turbidity (NTU) 70 – 0.73 Secchi depth (m) 70 + 6.5 TP (mg P m–3) 70 – 17.30

Note: Split sign (+ or –) indicates the direction of sorting; if negative, high values are sorted to the right, and if positive, high values are sorted to the left. The split constant is the cut-off point in the predictor variable for separating the two groups.

Fig. 1. Relationships between lake characteristics and instantaneous growth rate at 480 mm (mean length reached at two years old) of stocked rainbow trout in deep (≥20 m mean depth; ●) and shallow (≤11 m mean depth; ○) lakes in the North Island of New Zealand. Spearman rank correlations (rs) are given; *: P < 0.05, **: P < 0.01.

Fig. 2. Relationships between lake characteristics and instantaneous growth rate at 480 mm (mean length reached at two years old) of stocked rainbow trout in deep (≥20 m mean depth; ●) and shallow (≤11 m mean depth; ○) lakes in the North Island of New Zealand. Correlations are between growth parameters and annual means of environmental factors integrated through the entire water column of each lake (where appropriate). Spearman rank correlations (rs) are given; *: P < 0.05.

Fig. 3. Regression tree classifying trout growth rate on the basis of lake stratification and water quality variables, with scatterplots showing the spread of data (circles) and cut-off point separating groups (dashed lines). Decisions are given as inequalities, with the direction of sorting given by arrows. Mean growth rate (mm yr–1) in the resulting terminal nodes is given below group numbers. Boxes below group numbers show the group membership of measurements from different lakes and years; where no years are given, all measurements from that lake fit were classified into one group.

Supplementary Information Table S1. Characteristics of returned stocked rainbow trout used in comparisons of growth rate and L∞ among lakes. Values presented as means for 2002–2009. The number of fish returned by anglers is also given, as is the percentage of fish from each lake caught in each season.

Number of fish

returned Length of returned

fish (mm) Season of capture (percentage of total catch)

Lake Mean SE

Mean SE

Max. age of

returned fish

(years) Autumn Spring Summer Winter Spring releases

Okareka 33 4.80 274 7.87 4.59 30 14 26 31 Okataina 48 4.95 326 7.97 4.86 44 19 18 20 Rerewhakaaitu 35 4.79

280 6.75 4.44 41 20 16 27

Rotoehu 41 8.10 271 7.86 4.14 29 29 18 27 Rotoiti 42 3.35 304 7.47 3.72 36 25 22 18 Rotoma 21 3.40 258 7.15 5.00 28 23 32 26 Rotorua 43 5.69 290 7.46 3.88 33 17 33 20 Tarawera 55 5.83 320 4.01 4.49 34 23 26 18 Tutira 27 4.91 265 8.01 4.07 23 27 8 45 Spring Total 38 1.99

288 2.49 4.36 34 22 23 24

Autumn releases

Okataina 35 6.88 283 4.59 4.66 52 26 19 17 Rotoiti 38 4.72 324 5.73 3.48 32 32 19 21 Tarawera 45 6.60 299 5.06 4.50 31 20 29 20 Tutira 19 4.71 220 5.44 4.13 23 25 16 56 Autumn Total 34 3.21

283 2.93 4.17 31 24 19 26

Overall Total 37 1.70

286 4.31 34 23 22 26

Table S2. ANCOVA of log10-transformed rainbow trout weight among lakes, with log10-transformed fork length as a covariate. Rotoiti: n = 892, Rotorua: n = 4711, Tarawera: n = 265, Okataina: n = 319. Tukey’s HSD test: Rotoiti > Okataina > Rotorua > Tarawera; P < 0.001 for all comparisons.

df SS F P Intercept 1 62.09 21407 <0.001 Log10(length) 1 66.46 22914 <0.001 Lake 3 1.52 174 <0.001 Error 6182 17.93 Total 6186 117.10

Table S3. Spearman rank correlations (r) between rainbow trout growth parameters and environmental variables in nine shallow and deep lakes in New Zealand during 2002–2009. We calculated growth rate at 480 mm (mean length reached at two years old), and L∞ is a parameter of Bertalanffy’s growth equation. * indicates significance at P < 0.05, and ** indicates significance at P < 0.01. Variable (units) Growth

rate L∞ n

Lake morphometry      Catchment agriculture area: lake area -0.619 -0.500 8 Catchment forest area: lake area 0.762* 0.667 8 Catchment area (km2) 0.190 0.214 9 Lake area (km2) 0.433 0.250 9 Ratio catchment area: lake area -0.567 -0.383 9 Lake volume (m3) 0.571 0.500 8 Mean depth (m) 0.667* 0.600 9 Maximum depth (m) 0.800** 0.700* 9 Lake stratification  Summer thermocline depth (m) 0.700* 0.633 8

Hypolimnetic volume (m3) 0.762* 0.667 8

Epilimnetic volume (m3) 0.286 0.214 8 Epilimnetic volume: hypolimnetic volume -0.714* -0.619 8 Habitat suitability Mean water temperature (ºC) -0.567 -0.567 9 Mean DO (mg L-1) -0.595 -0.524 9 Minimum volume <21ºC (m3) 0.381 0.286 8 Minimum volume >6 mg L-1 DO (m3) 0.381 0.310 8 Minimum volume with suitable DO and temp (m3) 0.714* 0.595 8 Stocking rate (fish ha-1 yr-1) -0.190 -0.095 9 Trophic state TN (mg N m-3) -0.733* -0.667* 9 TP (mg P m-3) -0.283 -0.283 9 Chl a (mg m-3) -0.483 -0.467 9 Turbidity (NTU) -0.567 -0.533 9 Conductivity (µS cm-1) -0.214 -0.286 9 Secchi depth (m) 0.617 0.567 9

Fig. S1. Study areas and lakes in the North Island of New Zealand: (a) location of most study lakes; (b) location of Lake Tutira.