Stream temperature modeling in mountainous environments

Stream temperature is one of the key variables affecting the habitat suitability of numerous aquatic species. Over the past decades, research efforts on this topic have concentrated on low-land rivers of North-America, whereas mountainous environments have received much less attention - above all in Europe. The present thesis introduces two new models for stream temperature prediction in Alpine watersheds. Both are tested over selected catchments in Switzerland, a mountainous country which presents the advantage of possessing a dense network of automatic stream temperature measurement stations. The first model is specifically designed to provide stream temperature estimates in ungauged catchments, so as to compensate for the scarcity of temperature measurement sites in mountainous environments. Its design is based on a new statistical approach. As opposed to standard statistical models, which are common to many disciplines, the present one aims at incorporating some of the physics controlling stream temperature in its own structure. Its formulation is derived from an analytical solution to the equation describing the energy balance of an entire stream network. Some terms of this solution cannot be readily determined based on data available at the regional scale; they are approximated using standard statistical techniques. The resulting model is statistical in nature, but includes elements of thermodynamic principles. Its accuracy is shown to be similar to the one of a standard statistical model, its root mean square error being 1.3°C at the monthly time scale. In virtue of its physical basis, the model can be used to investigate into more detail the factors controlling stream temperature at the regional scale, as shown through a simple example. The second model is intended to provide deterministic stream temperature predictions, to be used for example in climate change studies. It builds upon an existing physically-based model, which has been entirely written anew in order to clarify its structure and ease future developments. Conceived as an add-on to the spatially distributed snow model Alpine3D, it simulates the flows of both water and energy within the catchment, based on a semi-distributed approach. Some components of the model can be represented using various alternatives; for example, three different techniques are available to simulate the temperature of subsurface runoff. This flexibility allows the model to be tuned to the specific needs of each user, but also permits a more thorough assessment of the simulation uncertainty by comparing the predictions of the various alternatives. Evaluation of the model in a high alpine watershed indicates that hourly mean discharge is reproduced with a Nash-Sutcliffe efficiency (NSE) of 0.82, and hourly mean stream temperature with a NSE of 0.78. Both models are shown to contribute to a better understanding of stream temperature dynamics in Alpine environments. Future work involves further research on the structure of the statistical model, as well as the application of the deterministic model within the framework of climate change studies.