Friday, 9 January 2009

An alternative index of rainfall that might be more helpful to real Australians

In a 2004 article about rainfall trends in Australia the CSIRO, who have been unfortunately but unsurprisingly exposed as greenhouse sceptics under the control of the coal and aluminum industries, made a very interesting suggestion to deal with the problem of rainfall totals failing utterly to represent the weather directly experienced by urban Australians and to a lesser extent farm conditions over the major agricultural areas of the continent.

In this study, Ian Smith, one of the premier climate-change scientists of the CSIRO, realised that averaged raw totals across Australia are not equally representative of all areas of the continent and believed that there must exists a better method of measuring average rainfall over Australia than the raw averaged totals, which he intuitively thought would favour the wettest areas. His solution was to averaged the mean decile range of rainfall over the continent. “Decile range” refers to the relative ranking of an annual or monthly rainfall total, with the lowest ten percent forming decile 1, the next lowest decile 2, and so on until the highest ten percent form decile ten.

Those with knowledge of rainfall over Australia will realise that in the most variable parts of the continent and generally in the drier months of the year rainfall distributions are highly skewed and the mean rainfall exceeded in only a small proportion of years. For example, in Derby, the mean May rainfall in 22 millimetres but the median only a fraction of a millimetre, with the 90 percentile (exceeded on average every ten years) being 65 millimetres.

Decile ranges thus show how many or months are actually wetter than the one being reviewed much better than departure from the mean. A graph comparing raw totals and mean deciles to measure rainfall for Australia can be seen here.

If one analyses this data, what one actually finds is that the raw totals do not necessarily reflect the wettest areas, but those with most variable rainfall. This should be obvious because their high deciles are more inflated in raw total than areas of low variability like western Tasmania, southeastern South Australia or southwestern Western Australia.

When one compares the rankings of raw totals and mean deciles, one sees that the year (from 1901 to 2002) which become most drier when mean deciles replace raw totals is 1977 (36th wettest by raw totals; 44th driest or 59th wettest by mean decile).

This year had very active monsoons in the Lake Eyre Basin but major rainfall failures in southern Australia outside of Western Tasmania. 1977 is unrivalled in this respect: its closest rival being 1959 (34th driest by raw totals and 18th driest by mean decile), which had severe drought over most of arid Australia, but very heavy rainfall in humid coastal New South Wales (which has for a region not affected by ENSO an amazingly variable rainfall that I have never understood).

In contrast, the year that becomes wetter to the largest extent from mean deciles replacing raw totals is 1966 (27th driest by raw totals; 47th driest by mean deciles), followed by 1946 (26th driest and 42nd driest), 1952 (16th and 30th driest) and 1958. 1952, of course, featured the weakest tropical wet season since 1885 and some of the most powerful southern storm systems ever resulting in the fourth wettest year since 1885 over Victoria and third wettest in Canberra. 1958 and 1966 saw similar wet season failures and notably heavy rain in at least parts of arid southern Australia. 1946, with its remarkably powerful mid-latitude winter westerlies, saw record dryness in northern New South Wales at the same time as Tasmania had one of its wettest years on record.

I also tried the experiment of comparing the standardised anomalies of raw totals and mean deciles (as shown by the thin grey line on the graph above). The results are somewhat different from the comparative rankings of raw totals and mean deciles, but are rather misleading because of the limits to standardised anomalies of mean decile values – which of course cannot be less than 1 or greater than 10.

The very fact that the ranking by mean deciles serves to devalue the remarkably infertile and sparsely populated tropical regions of the continent where rainfall can be very variable due to cyclones is enough reason to seriously test it. I would not dismiss possible applications to other regions of the world.

No comments: