Forecasting global climate change - Climate Change: The Facts - page 128

impending new ice ages that have occurred over the last 100 years and longer, including
those mentioned above. 28 And the rate is arguably consistent with the understanding of
scientists who consider that the Earth is still experiencing a cooling period, albeit with
fluctuations, that commenced around 4,000 years ago. 29
For horizons from one to 100 years from the year 1851 to the year 1975—7,550
forecasts in total—the average absolute errors of the 0.03°C per year warming forecasts
and of the 0.01°C per year cooling forecasts increase as the forecast horizon increases
(see Figure 1). Because our tests use historical data known to exhibit a warming trend,
the warming model has an unfair advantage in this test. Despite that advantage, across all
forecast horizons, the average errors of the warming forecasts are more than twice as large
as the errors from the relatively more conservative cooling hypothesis. Remarkably, the
natural cooling forecasts are more accurate than the dangerous warming forecasts for all
forecast horizons.
The global warming and cooling hypotheses were developed without the aid of
scientific forecasting. To develop a credible forecasting method against which to
benchmark the warming and cooling hypotheses, we needed a model that was both
consistentwithevidence-basedforecastingprinciplesandwithevidenceonclimatechange.
With that in mind, we asked climate expert and astrophysicist Soon to collaborate with us
to develop a model and validation tests. 30
With Soon, we established that the state of knowledge about the causes of climate
change was such that it would be inappropriate to develop a causal model. The strength
andevendirectionofproposedcausalrelationships,includingwithCO 2 ,aremuchdisputed
among leading climate scientists. 31 For example, in 1972 Kukla and Matthews reported
from a meeting of climate scientists that :
one conclusion reached at the session was that there is no qualitative difference between the climatic fluctuations in the
20 th Century and the climatic oscillations that occurred before the industrial era. The present climatic trends appear to
have entirely natural causes, and no firm evidence supports the opposite view. 32
A more recent analysis of two 3,000-year temperature proxy series comes to the same
conclusion. 33
We concluded from forecasting principles that because knowledge about climate
change is so poor, forecasts from a no-change forecasting model would be more accurate
than forecasts from methods that attempt to incorporate knowledge that is tentative at
best. Depending on the situation, the appropriate no-change model might be one that
forecasts that the level (e.g. current temperature) will not change, that the trend will not
change, or even that the rate of change will not change. For forecasting long-term global