Multi-scale streamfow simulation

Abstract

Scale issues represent an unsolved problem in hydrological sciences. Distributed hydrological models are capable of accounting for catchment heterogeneity, but it remains unclear to what extent the variations in the representation of spatio-temporal resolution in these models leads to uncertainty in the simulations. Moreover, the added value of more refned spatio-temporal discretization is also unclear. The present thesis addresses these topics in the context of streamfow simulation, food projection and regionalization of model parameters . All catchments studied in this thesis are located in Southern Quebec, Canada . This research uses two process-based distributed hydrological models with diferent degrees of complexity (Hydrotel and WaSiM). For the food simulation and projection, the models are calibrated with four diferent levels of spatial discretization of physiographic data and in models’ parameters. The climate extreme project (ClimEx) dataset is bias corrected for 3- and 24-hour time-steps using the n-dimensional multivariate bias correction method (MBCn), and used as inputs to the hydrological models to project streamfow over the 1991-2100 period. The results show that the variation of temporal resolution has only minor impacts on the uncertainty of historical simulations, and the impact depends on the choice of the model. The more sophisticated model (WaSim) has a larger uncertainty. As for varying the spatial discretization, it can cause uncertainties for catchments with low slopes or uneven areas. Regarding food projection, by refning the temporal scale, the results show that both the frequency and amplitude of extreme summer-fall fow increases in the future. Moreover, the choice of hydrological model for food projection is more important for larger catchments. Finally, no distinct pattern exists regarding the uncertainty related to the spatial resolution and catchment size. However, this afects the direction and signifcance of the trends observed for extreme fow in the simulations. This thesis also proposes and tests a regionalization method based on random forests (RF). It is applied to the parameters of Hydrotel at diferent spatio-temporal resolutions. The results show that the proposed regionalization technique performs better for shorter time-steps. Moreover, the regionalized parameters are spatially consistent. In the end, using catchment descriptors that have a better spatial representativity results in an improvement (more than 10%) in the simulations using a 24h time-step.

Date	27 Apr 2023
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Annie Poulin (Supervisor) & Marie Amélie Boucher (Co-supervisor)