‘Comment on “homogeneity analysis of rainfall series-an application of the use of a realistic rainfall model” ’

✍ Scribed by Thompson, C. S. ;Revfeim, K. J. A.

• Publisher: Wiley (John Wiley & Sons)
• Year: 1985
• Weight: 199 KB
• Volume: 5
• Category: Article
• ISSN: 2314-6214
• DOI: 10.1002/joc.3370050510

✦ Synopsis

We are grateful to Dr Jones for his commentary (Jones, 1985) and the opportunity to reply. Dr Jones has drawn attention to apparent flaws in the methodology presented by Thompson (1984). However in the context for which the methodology was advocated he may not have carried out similar tests on the 'tried and trusted' method(s). Shortcomings in any new methodology are to be expected as refinements a 2 pursued and are usually ironed out through a sequence of reporting stages.

1. The first basic flaw perceived by Dr Jones is that the parameters of the rainfall process may vary with t,me. They certainly vary seasonally as indicated by Revfeim (1982) and Thompson (1984). In the latest refinement a 6-harmonic model can be made to fit exactly the monthly means, albeit with 24 parameters! Nevertheless this joint estimation over all months does smooth out parameter variation compared with the analysis of individual months.

For long data series estimation of the physically meaningful parameters on a decade basis has not shown many statistically significant differences in the means, although the amplitudes of the first or second harmonics may differ. Hence the description that the underlying process was 'stationary' was perhaps a poor choice of terminology, since stationarity assumptions were not employed in the analytical procedures. Even stationary processes can produce a spurious 'decline' over several decades which could be underlying Sahelian rainfall.

However in Thompson's (1984) paper the likelihood statistic did compare both parameters of the current site against the test site. In effect this checks for a change in the mean rainfall since pp = f . The initial aim was to test for a change in the , M parameter alone, allowing p to be representative of the rainfall process over the separate time periods. A suitable test for this has yet to be devised.

By comparison the conventional methods with sequences of ratios between rainfall at two sites (usually within reasonable proximity) impose a different sort of constancy if these are taken to be from the same distribution over different time periods.

2. Of the four United Kingdom data series used to test the first version of the new methodology, only one, Stonyhurst, can be described as 'known' to be homogeneous since the original data series are unaltered by any homogenizing method. Presumably the site is unchanged, otherwise it would have been more sensible to also check for significant differences before and after site changes. However, we accept that spurious inhomogeneities have been identified by analysing individual months. Whether these differences would still show up after using a smoothed harmonic analysis, it is not known. When a test is devised maintaining representative estimates of rainfall recurrence for the different periods, few if any spurious inhomogeneities should be found.

3. It was never the intention in Thompson's paper to explain why there were certain inhomogeneities in some months and not in others, but merely to describe fully a new homogenizing process. The paper can be accepted or rejected on that basis. However, the implications of seemingly large adjustments in adjacent months is a question which is difficult to resolve. We do not have any specific explanations, but