SEARCH

SEARCH BY CITATION

Keywords:

  • artificial neural networks;
  • Atlantic salmon;
  • climate change;
  • ensemble;
  • thermal refugia;
  • water temperature

ABSTRACT

A stochastic model is proposed to reproduce daily water temperature at 18 observation sites (11 main stem and 7 tributary sites) in the Ouelle River basin located in southern Quebec, Canada, using meteorological variables as predictors. A random sampling procedure without replacement was adopted for the model calibration and validation to overcome the limited length of the observed water temperature series. The predicted water temperature series were then submitted to variance inflation to reproduce the observed variability of the water temperature series. Historical water temperature series were obtained from observed meteorological predictors, whereas reference and future water temperature series were obtained from stochastic water temperature model using five reference (1970–1999) and future (2046–2065) meteorological predictors simulated by five different climate model runs. The reference series reproduced summer mean water temperature and the number of consecutive days with water temperature higher than 21 °C or 25 °C fairly well. On the basis of the historical series, it can be assumed that the seven tributaries of the Ouelle River provided thermal refugia for native salmon between 1970 and 1999. Future water temperature series projected by the stochastic model show that the seven tributaries could still be used as refugia to prevent lethal stress, whereas the temperature in the main stem and in three tributaries will be high enough to constitute stressful conditions for feeding juvenile Atlantic salmon. Copyright © 2012 John Wiley & Sons, Ltd.