The Separation Effect and the Wakeby Distribution

Houghton argues that "It is possible to fit a single distribution, the Wakeby, with many combinations of parameter values that are easily compared, rather than several distribution functions, each with a different analytical form. The Wakeby distribution is a grand parent."

Moreover:

In traditional estimation procedures, the smallest observations can have a substantial effect on the right-hand side of the distribution. But the left-hand side does not necessarily add information to an estimate of a quantile on the right-hand side. 

In fact

Indeed, since floods are not known to follow any particular distribution, it seems intuitively better to divorce the left-hand side from the right... the Wakeby does exactly that.

The separation effect:

Matalas et al. (1975) [sampled] most of the commonly used distributions, repeated the plots for 10-year and 20-year records, and found that none of the distributions could reproduce as high a standard deviation as that found in nature. This has been termed the "separation effect". Thus, nature has skews that are even more unstable than those generated by common distributions. Moreover the authors showed that this separation effect cannot be explained by small sample properties or by auto-correlation.

And so:

The Wakeby distribution was originally introduced to account for this effect.

Moreover Kundar and Chander, p. 28 in Singh Ed.

The right the left hands of distribution can be modeled separately. The parameters a and b govern the left hand (low flows) tail while the parameters c, d and e govern the right hand (high flow) tail

References

Houghton () HERE