Water temperature modeling in the Garonne River (France)

Stream water temperature is one of the most important parameters for water quality and ecosystem studies. Temperature can influence many chemical and biological processes and therefore impacts on the living conditions and distribution of aquatic ecosystems. Simplified models such as statistical models can be very useful for practitioners and water resource management. The present study assessed two statistical models – an equilibrium-based model and stochastic autoregressive model with exogenous inputs – in modeling daily mean water temperatures in the Garonne River from 1988 to 2005. The equilibrium temperature-based model is an approach where net heat flux at the water surface is expressed as a simpler form than in traditional deterministic models. The stochastic autoregressive model with exogenous inputs consists of decomposing the water temperature time series into a seasonal component and a short-term component (residual component). The seasonal component was modeled by Fourier series and residuals by a second-order autoregressive process (Markov chain) with use of short-term air temperatures as exogenous input. The models were calibrated using data of the first half of the period 1988–2005 and validated on the second half. Calibration of the models was done using temperatures above 20 °C only to ensure better prediction of high temperatures that are currently at stake for the aquatic conditions of the Garonne River, and particularly for freshwater migrating fishes such as Atlantic Salmon (Salmo salar L.). The results obtained for both approaches indicated that both models performed well with an average root mean square error for observed temperatures above 20 °C that varied on an annual basis from 0.55 °C to 1.72 °C on validation, and good predictions of temporal occurrences and durations of three temperature threshold crossings linked to the conditions of migration and survival of Atlantic Salmon.