Simulating Streamflow In Manitoba Using North American Regional Reanalysis As Input Data
This study investigates the applicability of temperature and precipitation data from the North American Regional Reanalysis (NARR) for hydrological modelling of selected watersheds in northern Manitoba. For the specific region, it is found that NARR temperature and precipitation data are in much better agreement with observations than a popular global reanalysis data set. The hydrologic model SLURP (Semi-distributed Land Use-based Runoff Processes) was set up and calibrated for three catchments (Burntwood, Taylor, and Sapochi), using meteorological data from weather stations. When the calibrated models were run with temperature and precipitation data from NARR, runoff was underestimated by approximately 20 %. The SLURP model was then recalibrated using the NARR temperature and precipitation data as input. This eliminated much of the bias and provided a goodness-of-fit that was only slightly inferior to simulations with observed weather data. This suggests that SLURP can be adequately calibrated with NARR data and used for modelling hydrological processes in northern Manitoba where weather stations are scarce.
Comparison of Regionalization Methods for Flow Regime Simulation at Ungauged Basins in Ontario.
In this study, regionalization, a process of transferring hydrological information from gauged to ungauged basins, is used to simulate continuous flow regime in different watersheds across Ontario climatic regions. The entire study area covers approximately 1 million km2 and most of the basins have incomplete or short period of data records and most of them are located in the northern regions. Various regionalisation approaches have been proposed in the literature; however, it is unclear which methods could be the most appropriate for a given climatic and physical environment. The data collected for this study includes the catchment attributes (e.g. soil and geology types, landscape properties, shape characteristics of the catchments, etc) and computed model parameters of the integrated hydrologic modeling system (IHMS-HBV) for gauged basins. The regional model parameters used to simulate continuous flows in ungauged basins are obtained using a physical similarity approach through clustering technique, a non-physical similarity approach such as multiple regression, and methods which do not consider the role of catchment attributes such as kriging and weighting based on the inverse distance. Preliminary results based on jackknife cross correlation validation show that the weighting approach based on the inverse distance technique produces better results than kriging, multiple regression and clustering. Although kriging and multiple regression are the most largely used regionalization methods, they appear inappropriate in this region. This may be due to the large size of the catchments and/or the large number of selected attributes. Other emerging regionalization methods such as logistic regression implemented through neural network technique are under investigation. The best methods identified will be used to simulate flow regime at gauged and ungauged basins across Ontario. The flow regime information is essential to establishing environmental flow policy, regulations, and sustainable water management.
A Methodology for Calibrating a WATFLOOD Model of the Upper South Saskatchewan River
The upper South Saskatchewan River consists of the Red Deer River, the Bow River, and the Old Man River. With a contributing area of 120,000 km2, these three watersheds flow through a diverse range of land types including mountains, foothills and prairies. Using WATFLOOD, a model has been developed to simulate stream flow in this basin and this model is used as the case study for a straightforward calibration approach. The input for this model is interpolated rainfall data from twenty-three rain gauges throughout the basin, and the model output (stream flow) will be compared to measured stream flow data from thirty stream gauges. The basin is divided into nine land classes and four river classes. Because of the diversity of land types in this basin, proper identification of the parameters for individual land classes and river classes contributes significantly to the accuracy of the model. Critical land class and river class parameters are initially calibrated manually in representative sub-basins (comprised of >90%) of a single land class to determine the effect each parameter has on the system and to determine a reasonable starting estimate of each parameter. Once manual calibration is complete, DDS (Dynamically Dimensioned Search Algorithm) is used to automatically calibrate the model one sub-basin at a time. During this process only the parameters found significant during the manual calibration are altered and focus is on the land classes and river classes that dominate that sub-basin. The process of automated calibration is repeated once more but is done with multiple sub-basins and uses a stream flow weighting method. This is the final step towards a model that is calibrated to represent the diversity of the entire basin. The technique described is intended to be a general method for calibrating a regional scale model with diverse land types. The method is straight forward and allows adjusted parameters to provide relative accuracy over the entire basin.
Spatial patterns and temporal trends in hydrological regimes across the Americas
Land-use and water-use practices within the watershed can significantly affect water quantity and, as a result, instream ecological communities. In a changing climate, there is a clear need to understand the potential ecological implications of water quantity management. Climatic, GIS and hydrological data were extracted for watersheds along a north-south transect of countries in the Americas between 1971 - 1990 (maximum spatial coverage). Thirty-two ecologically-important variables from the Indicators of Hydrologic Alteration were calculated to characterize the variability between the catchments in the magnitude, frequency, duration, timing and flashiness (variability) of the long-term and annual hydrological regimes. Spatial trends in the hydrological quantity are explored at different scales through separate hierarchical agglomerative classification of long-term and annual hydrological regimes. For example, six long-term regime groups were identified reflecting the timing of annual peaks and low flows while four long-term regime groups were recognized reflecting the magnitude of the hydrological regime. Implications for changes to these hydroecological patterns in a changing climate are discussed as the results reinforce the importance of understanding spatial variability in water quantity as a precursor to sustainable water management.
Impacts of Operator Order in Hydrological Models
Operator-Splitting errors are inherent in many hydrological models and can lead to computational inefficiencies, parameter estimation issues and inaccurate model results. A detailed study of the operator splitting errors produced by standard implementations of VIC and TOPMODEL has been performed both to assess their significance and to evaluate methods to correct them. Both VIC and TOPMODEL have been incorporated into RAVEN, an object-oriented hydrological model developed at the University of Waterloo. RAVEN has been developed to model hydrological processes using flexible numerical algorithms. RAVEN's structure is specifically designed to separate the numerical methods from the conceptual design to provide flexibility and to allow a separation of the numerics from the physical representations. The advantages of this numerical and conceptual separation has led to a better understanding of the impacts that operator splitting has on existing hydrological models and has allowed for improvements within models that increase accuracy and minimize errors.