Étude de méthodes de régionalisation des paramètres des modèles hydrologiques et application à un bassin versant non-jaugé au Mexique

Abstract

The current project aims to determine and recreate historic flows on a Mexican basin that is not equipped with a hydrometric gauge. Such catchments, referred to as ungauged catchments, are impossible to model adequately because of a lack of reference to calibrate hydrologic models. To overcome this problem, methods to estimate streamflow on ungauged sites were developed. These methods, aptly named “regionalization” methods, attempt to transfer information from gauged sites in the same region to the ungauged site in order to estimate streamflow values. Four methods are typically used and provide different results. Unfortunately, no single method is consistently better than the others on all sites. Therefore this study compares three of the four methods which are based on hydrological model parameter transfer. The fourth method, based on multiple linear regressions, was not considered due to being historically inaccurate in conditions facing the catchment in this study. Since, to the author’s knowledge, no regionalization study has ever been performed in Mexico, the first part of the project was to apply and compare the regionalization methods in the study region, namely in central Mexico. This allowed finding the method which was most likely to succeed on the target site. The site is the Naolinco basin in Mexico which is a local and regional economic powerhouse and is currently subject to water quality and quantity problems. The findings of this research project show that regionalization methods in general are not applicable to catchments in Mexico, mostly due to the poor hydrometeorological network and data quality. Some regions were found to be more robust to regionalization, with the spatial proximity regionalization method slightly edging out the others and when multiple donor catchments were used. This holds true for all three models used (MOHYSE, GR4J and HSAMI). It is important to note that in cases where model calibration was successful, regionalization approaches were generally good. This shows that it is essential to carefully select the hydrological model and ensure that the model input data is as error-free as possible.

Date	21 Apr 2017
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Annie Poulin (Supervisor) & Richard Arsenault (Co-supervisor)