Geoscience Reference
In-Depth Information
AcompetingforecasterinAD115mightwellhavereasonedfromtheknowledgeofthe
time that our sun is like a large fire that must slowly burn down. Given that temperatures
over the previous century had been trending more upwards than downwards, she might
have proposed that while the Sun's fire may splutter and flare up from time to time, there
would be a long slow decline in the energy emitted. With these observations in mind, she
might have forecast that the average temperatures would trend downwards at the relatively
more conservative rate of 1°C per millennium or 0.001°C per year on average—a much
more conservative forecast than those of the first millennium warmer and of the twentieth
century warmers and coolers described above. While the errors of her cooling forecasts
increased only slowly out to the year AD 750, beyond that year the errors of her forecasts
tended to increase as the forecast horizon lengthened (dotted line in Figure 2).
We compared the records of the warming and cooling hypotheses forecasters with the
record of our benchmark no-trend hypothesis in the form of a forecaster who predicted
that the global average temperature for the 1,820 years from AD 116 to 1935 would be the
same as the AD 115 average. The solid line in Figure 2 shows the errors of the no-trend
forecasts by year, one forecast error per year. The modest size of the errors and the lack
of even a very small persistent trend in them suggest that there have been no changes in
theclimate system.Inotherwords,theclaim that'thingsaredifferentnow,'althoughoften
made in relation to forecasting in many fields, is once again unsupported. Over longer
policy-relevant periods, annual global mean temperatures are highly stable.
Even with a much more conservative forecast warming rate (onetenth that of our
previous tests), when applied to this series the warming hypothesis again performed
relatively poorly. The average error of the 1,820 years 0.3°C -per-century warming
forecasts was more than nine times the average error of the no-trend forecasts. Again, the
errorsincreasedwiththeforecasthorizon.Forexample,theerrorsofwarmingforecastsfor
the fourth century made in AD 115 were nearly three times larger than the errors of the
no-trend forecasts. The equivalent figures for the eighth, twelfth, sixteenth, and eighteenth
centurieswerefour,fourteen23,and27timeslarger.Thefindingsareconsistentwiththose
of Green, Armstrong, and Soon.
38
These findings from a long period of varied climate, then, are consistent with those of
our analysis for the 1851 to 1975 warming period above: the more conservative hypothesis
and forecasting method provides the more accurate forecasts. In particular, the most
conservativemodel,theno-trendmodel,hasgreaterpredictivevaliditythanlongtermtrend
models under diverse conditions. Nomatter when one starts forecasting and nomatter how
global average temperature is estimated, the evidence-based persistence model produces
by far the most accurate forecasts. The findings on the accuracy of forecasts from long
and short term tests of the alternative climate change hypotheses are summarised in Table
