Time filter

Source Type

Roy A.,Universite de Sherbrooke | Royer A.,Universite de Sherbrooke | Turcotte R.,Center dexpertise hydrique du Quebec
Journal of Hydrology | Year: 2010

Snow estimation is the major source of errors for spring streamflow simulations in Quebec, Canada. The objective of the study is to improve melting discharge estimation computed with the operational MOHYSE hydrological model by integrating remote sensing snow-cover area in its snow module (SPH-AV). The satellite-derived snow-cover area (SCA) (MODIS & IMS) is first compared with in situ snow depth data measurements and simulated snow-cover area. Results show that the remote sensing products underestimate the snow-cover area on the mainly forested study region. A direct-insertion method of daily satellite SCA images is developed based on an empirical snow water equivalent threshold compensating, on a pixel-by-pixel basis, for the small amount of snow that satellite sensors can not identify during the melting period. This approach improves the streamflow simulation for spring periods (25th March to 25th of May) over 4. years (2004-2007) with a Nash-Sutcliffe coefficient enhancement of 0.11 and a root mean square error (RMSE) improvement of 21% on the Du Nord watershed, for which the threshold was optimized. The threshold found on the Du Nord basin was then directly applied on another watershed (Aux Écorces basin) for validation. The simulated streamflow is significantly improved as compared to the observed streamflow for these 4. years (mean Nash increase from 0.72 to 0.85 and RMSE decrease by 22%).The method improves streamflow peaks identification as much as 36% on the Du Nord watershed and 19% on the Aux Écorces watershed. © 2010 Elsevier B.V.

Mailhot A.,INRS Eau | Lachance-Cloutier S.,Center dexpertise hydrique du Quebec | Talbot G.,INRS Eau | Favre A.-C.,Polytechnic School of Algiers
Journal of Hydrology | Year: 2013

The Peak-Over-Threshold (POT) approach is an interesting alternative to the one based on Annual Maxima (AM) series since it gives the opportunity to take into consideration extreme events that would not be considered otherwise. It has also been recognized that the regional approach improves statistical inference when compared to the local approach, assuming that the region is statistically homogeneous. A regional POT approach was developed and applied to the network stations located in southern Québec. POT series for 5-, 10-, 15-, 30-min and 1-, 2-, 6- and 12-h durations were constructed assuming a fixed exceedance rate. An analysis of local POT series showed that the intra-annual variability of the Generalized Pareto Distribution (GPD) parameters needs to be taken into consideration. Models of various complexities were defined combining local and regional representations as well as the intra-annual variability of GPD parameters. Regional likelihood was estimated and models were compared based on the Akaike Information Criterion (AIC). Models with regional shape and scale parameters and accounting for intra-annual variability were selected for all durations. Spatial covariates were also introduced through a simple model linking GPD parameters to latitude, longitude and altitude. The sensitivity of results to threshold values and selected models was also investigated. Interpolated maps of intense rainfall over the studied area are finally proposed. © 2012 Elsevier B.V.

Abaza M.,Laval University | Anctil F.,Laval University | Fortin V.,Recherche en Prevision Numerique Environnementale | Turcotte R.,Center dexpertise hydrique du Quebec
Journal of Hydrology | Year: 2014

This paper evaluates the application of the Ensemble Kalman Filter (EnKF) for streamflow assimilation within an ensemble prediction system designed for short-term hydrological forecasting at the outlet of the au Saumon watershed. The EnKF updates three state variables of a distributed hydrological model (soil moisture in the intermediate layer, soil moisture in the deep layer, and land routing) to improve the initial conditions of the forecasts. A systematic method for the identification of the perturbation factors (ensemble generation) and for the selection of the ensemble size is discussed. EnKF results show a substantial improvement in performance and reliability over the open-loop estimates. Manual assimilation was also assessed and led to a performance similar to the EnKF; however, the EnKF forecasts are substantially more reliable. While an ensemble size of 1000 members was required to fully sample the hydrological and meteorological uncertainty, similar results are obtained in terms of skill when limiting the ensemble size to 50. © 2014 Elsevier B.V.

Bergeron J.,Universite de Sherbrooke | Royer A.,Universite de Sherbrooke | Turcotte R.,Center dexpertise hydrique du Quebec | Roy A.,Universite de Sherbrooke
Hydrological Processes | Year: 2014

Estimation of the amount of water stored in snow is a principal source of error for spring streamflow simulations in snow-dominant regions. Measuring this variable throughout large and often remote areas using snow surveys is an expensive task since they are practically point measurements. Remote sensing is an alternative method, which can cover much larger areas in little time, but further research is required to reduce uncertainties on snow water equivalent (SWE) estimations, especially during the melting period. However, optical-near infrared (NIR) and passive microwave remote sensing can detect snow cover area (SCA) with greater certainty, which can be used as a proxy for SWE. The two datasets work in complementary ways considering their spatial resolutions and cloud cover limitations. This study developed an SCA product from blended passive microwave (Advanced Microwave Scanning Radiometer - Earth Observing System: AMSR-E) and optical-NIR (Moderate Resolution Imaging Spectroradiometer: MODIS) remote sensing data to improve estimates of streamflow caused by snowmelt during the spring period. The blended product was assimilated in a snowmelt model (SPH-AV) coupled with the MOHYSE hydrological model through a modified direct insertion method. SCA estimated from AMSR-E data was first compared with in situ snow-depth measurements and SCA estimated with MODIS. Results showed an agreement of over 95% between AMSR-E-derived and cloud-free MODIS-derived SCA products in the spring. Comparison with ground stations confirmed the underestimation of snow cover by AMSR-E. Assimilation of the blended snow product in SPH-AV coupled with MOHYSE yielded an overall improvement of the Nash-Sutcliffe coefficient comparable with simulations with no updates, which is comparable to results driven by biweekly snow surveys. Assimilation of remotely sensed passive microwave data was also found to have little positive impact on streamflow simulation due to the difficulty of differentiating melting snow from snow-free surfaces. © 2013 The Authors. Hydrological Processes published by John Wiley & Sons Ltd.

Abaza M.,Laval University | Anctil F.,Laval University | Fortin V.,Recherche en Prevision Numerique Environnementale | Turcotte R.,Center dexpertise hydrique du Quebec
Advances in Water Resources | Year: 2015

This paper evaluates Ensemble Kalman filter (EnKF) sequential data assimilation on a semi-distributed hydrological model implementation on two snow-dominated watersheds, focussing strictly on snow accumulation and melt periods while assimilating streamflow for the updating of various state variables combinations. Three scenarios are explored in depth: (1) updating the three state variables that were previously identified pertinent for snow-free hydrological processes: soil moisture in the intermediate layer, soil moisture in the deep layer, and the overland routing reservoir, (2) updating the snow water equivalent, and (3) updating all of the above state variables. Inputs (precipitation and temperature) and output (streamflow) perturbation factors are first identified for each scenario, based on their performance and reliability for simulation with assimilation. The three EnKF implementations are next compared to one another and to an open-loop run, in an ensemble forecasting context. The third scenario outperforms the others in most situations and provides the largest gain in reliability. The ensemble size may also be reduced, from 1000 to 50 members, without much loss in performance or reliability. © 2015 Elsevier Ltd.

Discover hidden collaborations