Statistical analysis
The proportion of perch in the fish community was calculated as the number of perch caught relative to the total number of fish caught in the littoral zone of the different lakes. To estimate how the proportion of perch in the catches changed over time we used a Linear Mixed Effect model (LME) with location (lakes) as random effect, using the nlme-package in R. The response variable was log(x+1) transformed. Relative density of fish in terms of Catch-Per-Unit-Effort (CPUE) was readily available for the two main study systems, Lake Skrukkebukta and Lake Vaggatem. The CPUE was calculated as the number of fish caught per 100 m2 per night (or 12 hours). To investigate the change in relative density over time, we estimated separate linear models for the two lakes, with log(CPUE+1) as the response and year as the predictor. To estimate the relationship between relative density and water temperature, we again used CPUE on a natural logarithmic scale with annual mean water temperature as predictor in a linear regression model. We transformed the predictor to a weighted-moving-average over the last two years with more weighting given to the latest year, to better reflect any long-term effects of temperature on relative density of perch. To investigate recruitment in the perch population, we used the relative proportion of one-year-old individuals within the perch population as a measure of the number of individuals surviving the first winter, which is suggested to be dependent on water temperature (Dahlke et al. 2020), but also food availability (or density of competitors) and density of predators their first growth season. Relative proportion of one-year-old individuals is calculated as the proportion of one-year-old perch relative to the total CPUE of the perch populations. We used linear regression with summer water temperature the preceding year on a natural logarithmic scale to predict the relative proportion of one-year-old individuals in the perch populations. In the statistical analyses, we combined both lakes as the number of sampling points were too low to treat them separately.