Statistical modelling of temporary stream flow in Canadian prairie provinces

Accurate forecasting of stream flow is of vital importance in semi-arid regions in order to meet the needs of humans, such as agriculture, and for wildlife. It is also of considerable interest for predicting stream flow for ungauged basins and for detecting change due to landuse or climate variations. Daily streamflows in semi-arid and arid regions are characterized by zero-inflation, seasonality, autoregression and extreme events such as floods and droughts. Analyses at the level of daily data for intermittent streams are problematic because of the preponderance of zero flows. Basic modelling approaches are often inappropriate when many zero flow events are present; approaches need to be modified to allow greater flexibility in incorporating zeros than is possible with traditional methods. This project discusses the utility of spline compartment models for analysis of data from intermittent streams, whereby the log-odds of the probability of a non-zero flow day, as well as the logarithm of non-zero flow rate can be studied. These models permit handling of large numbers of zero-flow days; the use of splines and other smoothers have the benefit that they permit a wide range of distributional shapes to be fitted. The models are illustrated for ten streams in the Canadian Prairie Provinces.