Modélisation de la température de l’eau dans une rivière à débit régulé en climats actuels et futurs

Abstract

The objective of this thesis is to further the understanding of water temperature modeling in regulated rivers in a context where hydropower and ecosystem management are priorities. The project was divided into three sub-objectives, each treated in a separate chapter of this work, all centered on the Nechako River located in British Columbia, Canada, and the water temperature simulated at the town of Vanderhoof downstream from a major spillway. In the first portion of the thesis, the applicability of the data is presented with an existing, well- known, deterministic coupled hydraulic and thermal model at an hourly timestep: HEC-RAS. Using this existing model structure, the observed meteorological data were removed and swapped with reanalysis data to determine the databases’ capacity to contribute meaningfully to the modelling process of the water temperature. Two portions of the reanalysis dataset were tested: ERA5 and ERA5-Land, whose data grid is more densely populated. Results were typically on par with what quality observed data was able to produce when using HEC-RAS on the Nechako River catchment, though the increased spatial density offered did not improve the robustness of the model. Of note, a thermal equilibrium was identified downstream from a series of lakes where the water resided for a few days near Cheslatta Falls. The upstream boundary condition, i.e., the water temperature at the outlet of the reservoir from which the river stems therefore has little impact on downstream temperatures. The next portion of the thesis builds on this work by applying the now calibrated HEC-RAS model with reanalysis data and uses it to study the impacts of climate change on water temperatures in the Nechako River. This was done by using 10 global circulation models from the CMIP6 ensemble whose available data corresponded to what the model required as inputs, paired with two shared socio-economic pathways representing varying environmental forcings and emissions scenarios (SSP2-4.5 and SSP5-8.5). The paper discussing this establishes that water temperatures are slated to rise over both the 2041-2070 horizon and the 2071-2100 horizon, more-so over the latter. Water temperatures are expected to increase upwards of 3 °C, which exceeds thermal limits commonly used for migrating species such as sockeye salmon and white sturgeon which may lead to increased mortality rates. The final portion of the thesis turns to a different model to simulate water temperature. Machine learning is leveraged through the use of a non-parametric empirical approach, a long short- term memory model, a type of recurrent neural network. Compared to a deterministic approach with a physically based model like HEC-RAS, an LSTM presents flexibility for its inputs and outputs. It has the capacity to use any data as input to train the model, including information from previous timesteps. The hyperparameters were calibrated through a random search and tested with various combinations of both observed (flow, water temperature at the boundaries, etc.) and reanalysis meteorological variables. The model performed well when using all available data, as expected. Though the required machine time for the calibrations was shorter than the HEC-RAS approach, it did require specific hardware and an increased knowledge of the software to initially set up, making this approach much less user-friendly. It does, however, provide an innovative and alternate avenue for modelling water temperatures to help validate and confirm existing findings and extend modelling to ungauged basins. Interestingly, variables like air temperature, expected to have the greatest influence on water temperatures, are not the most impactful. In the case of the Nechako River, lateral inflows from the Nautley River, one of the main tributaries, are much more influential in a sensitivity analysis than the meteorological variables identified in the study.

Date	25 Mar 2025
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Richard Arsenault (Supervisor) & André Saint-Hilaire (Co-supervisor)