Fitting Annual Floods with Zero-Flows

Many rivers in the dry zones of the Canadian prairies have zero flows in their annual flood records. These flood series can be treated by first estimating the probability of zero flow, and then fitting the non-zero flows with such methods as the Gumbel, Generalized Extreme-Value, Pearson Type III or Wakeby distributions, or Non-parametric Kernel estimation. Applying the five methods to several Canadian rivers, the Wakeby distribution and the non-parametric approach provide the best fit. These two methods are relatively sensitive to sample size, as is demonstrated by applying all five methods to short annual flood sequences simulated by bootstrap technique. For short records with many zero-flows, the Gumbel and the Pearson Type III distributions appear to be more appropriate.