Naive forecasts outperform IPCC

The three researchers cre­at­ed a naive ‘bench­mark’ fore­cast of tem­per­a­ture over the 16-year peri­od since 1992 (when the IPCC issued it’s 3° per cen­tu­ry fore­cast of glob­al warm­ing). Their naive fore­cast was that the tem­per­a­ture over the rest of the peri­od would be the same as the mean tem­per­a­ture in 1992. When com­pared with both HADCRUT (UK Met.Office) and UAH (Satel­lite) obser­va­tions, their naive ‘fore­cast’ per­formed as well as the UN’s elab­o­rate glob­al cir­cu­la­tion mod­els. When ‘back-cast’ against the HADCRUT series of tem­per­a­tures between 1850 and 1975—a year cho­sen to min­i­mize ‘cher­ry pick­ing’ the data—the naive fore­cast sig­nif­i­cant­ly out­per­formed the IPCC mod­eled forecasts.

The chart extract­ed from the Green, Arm­strong and Soon paper shows the use of the naive fore­cast alone as a ‘bench­mark’ fore­cast of tem­per­a­ture after 1850 (HADCRUT):

We used each year’s mean glob­al tem­per­a­ture as a fore­cast of each sub­se­quent year in the future and cal­cu­lat­ed the errors rel­a­tive to the mea­sure­ments for those years. For exam­ple, the year 1850 tem­per­a­ture mea­sure­ment from Hadley was our fore­cast of the aver­age tem­per­a­ture for each year from 1851 through 1950. We cal­cu­lat­ed the dif­fer­ences between this bench­mark fore­cast and the Hadley mea­sure­ment for each year of this 100-year fore­cast hori­zon. In this way we obtained from the Hadley data 157 error esti­mates for one-year-ahead fore­casts, 156 for two-year-ahead fore­casts, and so on up to 58 error esti­mates for 100-year-ahead fore­casts; a total of 10,750 fore­casts across all hori­zons.

[The chart] shows that mean absolute errors from our bench­mark mod­el increased from less than 0.1°C for one-year-ahead fore­casts to less than 0.4°C for 100-year-ahead fore­casts. Max­i­mum absolute errors increased from slight­ly more than 0.3°C for one-year-ahead fore­casts to less than 1.0°C for 100-year-ahead forecasts. ”

The con­clu­sion that the three fore­cast­ers reach is:

Glob­al mean tem­per­a­tures were found to be remark­ably sta­ble over pol­i­cy-rel­e­vant hori­zons. The bench­mark fore­cast is that the glob­al mean tem­per­a­ture for each year for the rest of this cen­tu­ry will be with­in 0.5°C of the 2008 fig­ure. There is lit­tle room for improv­ing the accu­ra­cy of fore­casts from our bench­mark model…our analy­sis shows that errors from the bench­mark fore­casts would have been so small that they would not have been of con­cern to deci­sion mak­ers who relied on them. ”

In the dis­cus­sion they note three prin­ci­ples for pol­i­cy fore­cast­ing that these results sug­gest must be applied to cli­mate forecasts: 

  • Val­i­da­tion tests should prop­er­ly be con­duct­ed on fore­casts from evi­dence-based fore­cast­ing pro­ce­dures. The mod­els should be clear­ly spec­i­fied, ful­ly-dis­closed, and replic­a­ble. The con­di­tions under which the fore­casts apply should be described. 
  • Spec­u­la­tion is not suf­fi­cient for fore­cast­ing. The belief that “things have changed” and the future can­not be judged by the past is com­mon, but invalid. 
  • Final­ly, suc­cess in fore­cast­ing cli­mate change and the effects of cli­mate change must then be fol­lowed by valid fore­casts of the effects of alter­na­tive policies.

Leave a Comment

Your email address will not be published. Required fields are marked *