Scientific and operational value of alternative datasets in hydrological science

Applied hydrological science strongly relies on hydrometeorological information, and especially on measurements of precipitation and temperature. However, time series of relevant hydrometeorological variables are plagued with problems such as short time series, missing data, errors, and various biases. This limits, sometimes severely, the ability to adequately represent the spatial and temporal variability of the hydrological processes at the basin scale. New datasets are proposed in this project to palliate the lack of direct observations. Available datasets consist of the traditional station-based data and of three alternative datasets: gridded datasets, reanalyses of observations and data generated from regional climate models. The scientists and industrial partners involved in this project propose to investigate the scientific and operational value of alternative datasets in hydrological science and to examine the projected changes on hydrology at the basin scale.

• Compare traditional and new datasets with respect to their ability to provide useful information for key hydrological applications and research.

• Further explore the advantages and drawbacks of observationally-based datasets.

• Estimate the added value of dynamically generated datasets in hydrological studies through the use of outputs from regional climate model simulations.

• Evaluate the impact of climate change on basin-scale hydrology through the use of high-resolution projections from the Canadian Regional Climate Model (CRCM).

The methodology is centered around five research themes that will compare the different datasets with respect to their ability at providing relevant information for hydrological modelling, extremes analysis, as well as provide insights on the optimization of observation networks and on the added value of complex physically-based hydrological models.

• Assess the reliability of both observed and simulated gridded datasets for the precipitation (P) and surface temperature (T) fields.

• Examine network rationalization through optimal configuration for P and T for hydrological modelling.

• Assess the impacts of using different datasets on hydrological modelling results.

• Evaluate the ability of a complex physically-based distributed hydrological model to incorporate high-resolution P and T, as well as additional meteorological inputs.

• Evaluate extreme precipitation from the different datasets to provide useful information.

This project evaluated gridded datasets interpolated from station data, atmospheric reanalysis and regional climate model data output for use in hydrological science. This project generated the first Canadian watershed database (CANOPEX). This database (available at canopex.etsmtl.net) contains data for 698 watersheds across Canada (Figure 1).

Figure 1 : Carte des 698 bassins versants inclus dans la base de données CANOPEX et leur précipitation annuelle moyenne tirée de la base de données de Ressources naturelles Canada (mm).

The key results from this study are as follows :

• Gridded datasets interpolated from station data all differ in various ways depending on the interpolation scheme. The differences are especially noticeable over regions with sparse station coverage. However, for general hydrological modeling purposes, all such datasets can be considered equivalent for simulating streamflows at basin outlet;

• Interpolating from station data results in a smoothing of extreme values. This did not cause problems for hydrological modeling performed at the watershed scale (for surface areas greater than 500 km2 in this study) but is a key limitation for studies in urban areas and small watersheds;

• Combining all datasets in a coherent manner resulted in better hydrological modeling than using any individual dataset alone;

• Modern reanalysis is as good as station-derived databases for hydrological modeling over North America, with the exception of the eastern USA where the annual precipitation cycle was more difficult to reproduce.

• The use of reanalysis as reference datasets has high potential, especially in the future, as the spatial and temporal resolutions will get finer, and as assimilation schemes become more sophisticated and integrate even more observations;
  Although results indicate that reanalysis adequately reproduced historical records and extremes, it seems necessary to post-treat these series before they can be used for the development of IDF (Intensity-Duration-Frequency) curves, for example;

• An analysis of the Canadian Precipitation Analysis (CaPA) dataset confirms the strong potential of using numerical model outputs as reference datasets. Results suggest that, in the future, the most reliable reference datasets will be issued from the combination of numerical models (analysis) and observations (ground and atmospheric);

• Outputs from high-resolution regional climate models remain too biased to be considered as reliable reference datasets, as they do not assimilate observations like reanalyses do. However, outputs from such models can be useful to test real-world hypotheses about key hydrological processes such as snowmelt;

• Reanalysis-driven CRCM precipitation structures displayed intensities spatially more homogeneous than observed ones for the central and eastern United States. Despite generating significantly lower precipitation volumes, intensities and area, they did reproduce annual cycles of characteristic values fairly well. Rainfall properties are also analyzed through the simple scaling theory that considered recorded series over Canada and USA. Preliminary results indicate that, globally, the relationships among extreme probability distributions at several durations are coherent over large regions.

All the above results are outlined in 16 refered articles and 18 conferences or posters. A total of 17 highly qualified personnel (HQP) were trained through this project.

• Institut national de recherche scientifique (INRS)

• Université du Québec à Chicoutimi (UQAC)