Regionalisation of skew for food frequency analysis : a case study for eastern NSW

In food frequency analyses, skew plays an important role in characterising the tails of the flood frequency distributions for which a three-parameter distribution like Log-Pearson Type 3 (LP3) is often preferred over two-parameter distributions. For fitting the LP3 distribution to a station's data, an estimation of skew is needed. Since the length of streamflow data is limited for the majority of stations in Australia, the at-site estimation of skew is highly uncertain. To overcome this problem, the use of regional skew has been advocated, however, this has not been well investigated with Australian data. This paper presents a Bayesian generalised least squares (B-GLS) regression approach to regionalise the skew coefficient for eastern NSW. It has been found that the B-GLS regression approach is quite capable of providing stable estimation of skew, which is equivalent to an at-site skew estimator based on over 90 years of data. The results of this study suggest that the B-GLS model can provide an avenue to develop a regional skew map for Australia using the national database being prepared as a part of the on-going revision of Australian Rainfall and Runoff.