Statistical Modelling of Temporary Stream Flow in Canadian Prairie Provinces
Accurate forecasting of streamflow is of vital importance in semi-arid regions in Canada in order to meet the needs of humans, including for agriculture, and of wildlife. Daily streamflow discharges in semi-arid and arid regions are characterized by zero-inflation, seasonality, autoregression and extreme events such as floods. Typically, flood frequency analysis estimates the level of the T-year flood based on a probability distribution model postulated for annual extrema. When many zero flow events are present, the postulated probability distribution should be modified using a mixture distribution for the zero component. While an analysis based on annual maxima avoids the need for modelling seasonal variation and serial correlation, such an approach requires very long time series. Furthermore, valuable information based on daily data is not being exploited. Hydrological issues of flow, in particular zero flow intermittent streams that make use of daily data have received little attention in the literature.
For this project, statistical models for the log-odds of the probability of a non-zero flow day and the logarithm of non-zero flowrate are studied, where seasonality and autoregression in the flow distributions are taken into account. The models are illustrated using several streams in the Canadian Prairie Provinces.