Evaluation of Statistical River Temperature Forecast Models for Fisheries Management

Abstract

AbstractWarming rivers and an improved knowledge of thermal impacts on fish are fueling a need for simple tools to generate water temperature forecasts that aid in decision making for the management of aquatic resources. Although there is strong evidence for temperature‐dependent mortality in freshwater fish populations, the application of water temperature models for in‐season fisheries management is still limited due to a lack of appropriate temperature thresholds and due to uncertainty in forecasts. We evaluated the ability of statistical models based on seasonal trends, air temperature, and discharge to produce daily forecasts of water temperature in the Fraser River, British Columbia, including explicit quantification of uncertainty in predictor variables. For all models evaluated (with and without air temperature and/or discharge predictor variables), the top model choice varied as a function of environmental conditions, uncertainty in the air temperature forecasts used to predict water temperature, and the selection of quantitative performance criteria (i.e., defining the “best” model based on the smallest mean raw error or based on the ability to accurately forecast extreme water temperatures). Water temperature forecasts averaged across 10 d produced by simple models that were fitted only to historical seasonal water temperature trends were as accurate as forecasts generated from uncertain air temperature predictions. Models fitted to air temperature were critical for forecasting high temperature thresholds; even the use of uncertain air temperature forecasts predicted high water temperatures with greater accuracy than models that lacked an air temperature covariate. In contrast, models that were fitted to discharge variables lowered the rate of false‐negative and false‐positive errors associated with estimating below‐average temperatures. On the basis of our findings, we suggest that fisheries managers should quantify the effect of uncertainties in model predictor variables when assessing water temperature models and should evaluate model performance in the context of system‐specific conditions and management objectives.Received May 15, 2013; accepted September 13, 2013Published online January 31, 2014