poly: Weather prediction falls as power law

Since people like to use the weather as a methaphor about our
difficulty in predicting the future, it was interesting to learn
that using simple correlations, predictability falls as a power
law with time, not an exponential. More sophisticated prediction
approaches should make the far future even more predictable
relative to the near future.

>THE PERSISTENCE OF WEATHER. Although there seems to be
>nothing reliable about the weather, especially when planning a
>picnic, there is actually a (very) long range consistency at work.
>Armin Bunde (armin.bunde@theo.physik.uni-giessen.de) at the
>University of Giessen, and his colleagues from Milan, Potsdam, and
>the Bar-Lan University (Israel) have now conducted the most
>thorough multi-year study yet of correlations in daily temperature
>records. What they do in effect is to ask: if the weather is sunny
>and warm today, what will be the likelihood that it will be sunny
>and warm tomorrow, and the next day, and after x day? Choosing
>randomly 14 meteorological stations worldwide---ranging from
>Pendleton, Colorado (with 57 years of data) to Prague (with the
>longest daily temperature record, 218 years)---and factoring out
>seasonal effects by comparing not temperatures but departures from
>the average daily temperature, they are able to tease out the
>persistent rhythms of temperature. As expected, they observed that
>after x days, the weather is less and less likely to be similar to that
>on day 1. In particular, they find (1) that the falloff in correlation
>is not exponential in nature (e to the -x power) as many had
>thought; rather it obeys a power-law rule (i.e., the correlation is
>actually proportional to x raised to some exponent); (2) the
>exponent, with a value of -0.65, is the roughly the same for all the
>cities; and (3) the persistence of this behavior seems to hold over at
>least a decade and maybe as long as a century or more. (Koscielny-
>Bunde et al., Physical Review Letters, 20 July; see fuller account
>at www.aip.org/physnews/preview.)

Robin Hanson
hanson@econ.berkeley.edu http://hanson.berkeley.edu/
RWJF Health Policy Scholar, Sch. of Public Health 510-643-1884
140 Warren Hall, UC Berkeley, CA 94720-7360 FAX: 510-643-2627
Received on Mon Jul 27 19:49:18 1998