The mathematical futility of long-term climate projections

The entire climate-prediction edifice as currently promoted, is founded on quicksand, as time-series research at LSE demonstrates - the Hawkmoth Effect rules. The issue comes down to several Inconvenient Modelling Facts:

• Complex systems do not generally have the mathematical property of Structural stability. This leads to quite different projections, using the same initial conditions, even if two models differ to a minute degree - the poster at the link graphically illustrates this.
• Models used for projections simplify real phyical processes (for tractability) and make assumptions (for tuning) - both of which cause structural inadequacy by incomplete description of the phenomena being modeled.
• Turbulent/chaotic systems are pointless to model over long time series if accurate projections are expected: each time-slice depends on the output of the former slice as initial conditions and projections thus rapidly depart from sensible physical-result boundaries as model-time progresses
• The real physical world, as best can be gleaned from observation of history, has close-to-absolute boundaries on temperature = +/- 2-3 degrees Centigrade, and two dominant Attractors: Ice Age and InterGlacial. There have been more than enough long-tail (extremely low- probability) perturbations in the past (vulcanism, comet strikes) to supply a record of any vast excursions from these boundaries. The bounds seem to hold, regardless
• The 'Escape from Model-land' LSE paper suggests that consistency with the past, use of long-tail (very low probability) inputs and expert judgement as to model applicability and utility in the first place, is the only way out of Model-Land
• "The utility and decision-relevance of these model simulations must be judged based on consistency with the past, and out-of-sample predictive performance and expert judgement, never based solely on the plausibility of their underlying principles or on the visual “realism” of outputs."

The two turbulent/chaotic systems which climateers attempt to model are, of course, Atmosphere and Ocean. 

Essentially, they are trying to predict the weather on 27 March, 2119, when predicting the weather on 27 April 2019 is clearly a nonsense: a coin-toss or a dartboard are more useful Models. Next weekend's weather - a reasonable prediction is likely. Tomorrow - the prediction will be close to spot-on. 

Models are always wrong, but mostly they are Useful.....except for the Far Future.  Mathematics rules.