Insights Applied to Climate Modeling.
Prediction of Monthly Global Temperatures.
We present here an updated validation of the initial ex ante prediction for May 2011 to October 2017 (78 months) of this system model by a comparison of observed temperatures (black/withe square dots; HADCRUT3) vs predicted temperatures (red lines). Both model and ex ante prediction have not been changed since their publication in 2011. However, the HADCRUT3 dataset, a joint data product of the UK Met Office Hadley Centre and the Climate Research Unit at the University of East Anglia, which was used for model building is no longer supported by the providers. Instead, a new version, HADCRUT4, is maintained now, whose values differ from the previous version ones. In order to keep our system prediction up-to-date, we now have to transform HADCRUT4 into HADCRUT3 values, which introduces minor deviations from the original HADCRUT3 data, however. Apart from that, this is real OUT-OF-SAMPLE prediction.
As of December 2016, the prediction accuracy of the most likely prediction (solid red line) of the Insights model is 59%. The accuracy relative to the prediction range (pink area) is 93% (fig. 1).
In comparison, the expensive General Circulation Models (GCMs) which the IPCC AR4 and AR5 projections are based on and which rely on atmospheric CO2 as major climate driver (and which are long-term trend models for 100 years, to be fair), show a prediction accuracy of just 31% for the time period 2007 (the year of publication) till today and 29% for the same time period as the Insights system model (fig. 2).
In contrast, there are satellite based temperature observation data of the lower troposphere of which the RSS (Remote Sensing Systems) and UAH (University of Alabama in Huntsville) datasets are the most prominent ones. They show higher spatial resolution, cover over 98% of the earth's surface, and they do not use interpolation techniques for spatial grid correction (though they have to apply calibration methods for consistency of the datasets), which improves data quality and reliability.
A comparison of the ex ante system prediction with these observed satellite data (average of RSS and UAH) shows an even higher forecasting accuracy of currently 61% and almost no bias of the most likely predictions, with the exception of the extreme El Niño weather event in 2015/16 (fig. 3). Also, it is obvious that the moderate IPCC A1B forecast is above the observed temperature data most of the time, and it overestimated the temperature development over the past 10 years, except for few months of the El Niños of 2009/10 and 2015/16.
The demonstrated high predictive power of the self-organized Insights system model is clearly a result of the ability of the implemented inductive modeling technology to autonomously and reliably extract relevant information from the noisy observational data for modeling the internal workings of the ill-defined climate system than theory-driven modeling approaches can achieve based on incomplete and uncertain human knowledge about the system. (Read more about inductive self-organizing modeling and ill-defined systems in the references included in the free download package.)
"The idea is to have the computer construct a model of optimal complexity based only on data and not on any preconceived ideas of the researcher; that is, by knowing only simple input-output relationships of the system, [the] algorithm will construct a self-organizing model that can be used to solve prediction, identification, control synthesis, and other system problems."
-- Stanley J. Farlow, 1981, in: "The GMDH Algorithm of Ivakhnenko."