NEW DEVELOPMENT IN WATERSHED MODELING USING MULTI-SITE GENERATION OF DAILY WEATHER DATA

Weather generators have been used successfully for a wide array of applications such as hydrology, agricultural and environment. Unfortunately, most weather models ignored spatial dependence exhibited by weather series at multiple sites because of climatic phenomena which extend over region rather than station location and constrain the observations in a given place to be correlated to those in surrounding area.

A stochastic weather generator was used to simulate precipitation and temperature data. Two approaches were used: uni-site and multi-site generation. The uni-site approach simulates the weather data at each station independently from the others. Thus, the spatial correlation in precipitation data is ignored. The multi-site approach produces spatial correlations that are identical to those observed. This approach is based on spatial autocorrelation to analyze pattern in space and investigate the dependence of weather data at multiple locations.

To test the effectiveness of multi-site approach, Peribonca River Basin in the province Quebec of Canada was used. The daily spatial dependence between precipitation occurrences and amounts and between temperatures was successfully reproduced as well as the total monthly precipitation, monthly numbers of rainy days and average monthly temperatures. A hydrological model was used to evaluate the differences between the two approaches using various criteria including annual runoff volume and annual and seasonal peak discharge. As envisaged, the multi-site generation of weather data greatly improved the simulation results and made the hydrological model more realistic and practical.