Updated February 20, 2014
"Science is the belief in the ignorance of the experts."
- Richard P. Feynman
Published in Synthetic Information May 11, 2013
Published in Synthetic Information October 8, 2013
Modeling of the sort used by the climate modelers, on the other hand, plays a rather minor role in the hard sciences. Modeling is generally viewed as a low level activity mainly useful for analysis of experimental data. It comes as no surprise to scientists that climate models are so often wrong, and so often modified.
INTRODUCTIONWe ask the question: Is the methodology of climate modeling fundamentally unsound? Further, we examine the track record of climate forecasting models. Over the past two decades substantial new climate data has become available from accurate sophisticated monitoring and imaging systems. We will highlight global sea ice measurements made by modern satellite imaging systems. Importantly, these new data were acquired after climate modelers made their predictions about sea ice. More generally, climate forecasting models are now subject to a host of new critical tests. Did their widely publicized predictions come true? How badly have climate model forecasts missed the mark?
In science, if model disagrees with new data in any significant way, that is, if a model forecast or prediction is contraindicated by new data, the model is invalid, or just plain wrong. Maybe the model can be fixed up, and maybe not. When model predictions exhibit significant long term divergence from new data, it's a strong indication that the underlying methodology of the model is fundamentally unsound. That's how the scientific method works.
In scientific modeling, new models are always suspect. Old models can be valid, but only if they are validated by new data. Further, valid old models that diverge from new data become invalid old models. The scientific method requires rather stringent conditions be met by models, and models are happily discarded if they don't make the grade. If something is learned from the erroneous model, we have scientific progress.
In climate modeling, new models are bad unless proved good by experiment. That is, new models have no scientific standing or claim to validity until validated by actual observed agreement with future data. The graveyards of science are filled by new models that subsequently failed to pass rigorous testing and repeated confirmation by experiments.
Should the Lemon Law Apply to Climate Models?When climate models are churned out annually, biannually, or multiannually, newly tweeked to fit the latest data, as new models, they have no validity. Model validity is only acquired if predictive accuracy is demonstrated over many future years, without further tweaking. Models that undergo periodic tweaking to agree with new data become little more than curve fitting exercises, no matter how fancy the graphics. This kind of ad hoc methodology amounts to little more than pure empiricism.
In science, old models may be valid or invalid. New models are always invalid (or unvalidated) until tested. They may become validated after they demonstrate substantial agreement with new data (i.e. data collected after the release date of the model in question.)
Climate modelers are in an embarrassing situation. Searching questions about the validity of the climate modeling industry are being asked. Is the climate modeling process fundamentally flawed? Is climate modeling a lemon?
When we purchase a new car that breaks down, and is generally unreliable during the next year after purchase or repair, the car is considered a lemon and can't be fixed. We have legal grounds to return the car to the manufacturer and get our money back.
If a pharmaceutical company comes up with a new drug, how do we consumers know it's effective and safe? We don't, nor does anyone else know. We only find out by clinical trials over many months or years that confirm effectiveness and safety.
Like new untested drugs, new climate models should be considered unusable until thoroughly tested and proved effective over time. If we hold climate models to this clinical standard, well...the old models didn't make it through clinical testing, and new ad hoc models probably won't either.
Our broker would say: It's probably time to sell climate modeling short. Take the loss, and get out of that position.
Recall that the proponents of climate models themselves made repeated extravagant claims of the predictive accuracy of their models. Extravagant claims of predictive accuracy must be backed up by extraordinary agreement with future data. For the climate modelers, a miss is as good as a mile.
Proxy time animations generate pretty pictures that have no causal interrelationship, and that's dangerous and (often) misleading. Such animations need not follow the actual time evolution of the real physical system being modeled.
Animations of an anvil falling on the Coyote do not accurately simulate a real anvil falling on a real Coyote.
Climate model animations using proxy time might appropriately be called Climatoons or Proxytimeatoons.
Proxy time Climatoons are not educational, cool, cute, or funny. They are misleading to the uninformed, and of little value to the informed.
Climate Models and Climate Forecasting
Over the past few decades substantial new climate data has been accumulated using sophisticated monitoring and imaging technology. This new and extensive climate data present more rigorous tests of the widely announced predictions of climate models. It is now clear that new climate data accumulated 1995-2013, when compared to predictions of pre-1995 climate models show that those models have often (nearly always) diverged significantly from the real climate.
In the following, we compare a few egregious climate model predictions to new data, and explain our view that the methodology used in climate models is inherently flawed.
Climate forecasting models and their failures.
What about sea level rise?
Here's the latest data, and it's really impressive.
What data is plotted in this graph?
Some observations about these data:
First, the maximum north polar sea ice coverage occurs in the month of March, and the maximum area of coverage ranges from roughly 10.5 to 11.5 million km^2. The minimum sea ice coverage occurs in the month of September, and the minimum sea ice coverage area ranges from roughly 2.5 to 4.3 million km^2. Every year the sea ice area cycles between these maxima and minima in March and September.
Second, we observe that the north polar sea ice coverage over 2013 (black curve) is generally higher or comparable to the average of years 2005 through 2012.
Third, in the month of October 2013 the measured arctic sea ice extent is relatively larger than all previous year's October coverage. That is, arctic sea ice reached record high levels in October 2013, according to month-on-month comparisons over available data from previous years (2005-2012.)
First, it is clear that many of the highly publicized IPCC climate model predictions about polar ice cap meltdown or steady year-on-year sea ice reductions did not occur.
Of course, models can be tweeked, modified or otherwise fixed up after they fail. When modelers do this in a way that agrees with the latest data, we have a new model. After such ad hoc fixes, the new models are considered invalid until they demonstrate significant long term agreement with new data, without further tweeking.
What is proxy time?
The parameter representing time, the time variable, does not appear in the physical theory of thermodynamics. One has thermodynamic state variables, but no time variable. Further, many climate simulation models do not support a time variable at all. For example, steady state flow models of the earth's ocean and atmosphere solve a set of equilibrium equations (with boundary) that do not have time as a variable.
So how can such models having no time variable, be used to predict the future time evolution of the system?
Many such models make use of an artificial time or proxy time. Some quantity that varies monotonically with time, is used as a substitute for the time variable.
For, example, during time periods where atmospheric CO2 concentration is increasing steadily (monotonically) with time, one can plot model output as a function of CO2 concentration. Some modelers convert from CO2 concentration to a proxy time variable, in order to plot the model output as a function of proxy time.
Successful Non-causal Models Exist and are Especially Useful for Describing Periodic Phenomena.
Interestingly some non-causal models can exhibit excellent long term predictive capabilities in a restricted range of model parameters for simple systems. For example, Newtonian gravitational field theory coupled with classical Newtonian mechanics is a non-causal model.
Why do we say Newtonian gravity is non-causal?
The Newtonian gravitational field equations are non-causal because they exhibit instantaneous action at a distance. There are no gravitational waves or propagation speeds for changes in the Newtonian gravitational field. General relativity on the other hand, does exhibit causality, there is no "action a a distance" in general relativity. Changes in the metric propagate at the speed of light in general relativity.
Further, Newtonian mechanics does not obey the rules of special relativity. Despite these problems, Newtonian mechanics can work quite well as a predictive theory.
For simple systems, e.g. planetary orbits and the like, standard Newtonian gravity and mechanics have excellent predictive capabilities and often show impressive long term agreement with measured motions of planets, asteroids, comets, slow speed space ships, etc. In many cases, the agreement with new measurements (data) extends over many years and many orbital periods. Importantly, periodic or quasi-periodic phenomena are often well described by non-causal models.
But Newtonian mechanics does not always work. The theory is contraindicated by measurements of time evolution of the perihelion of the planet Mercury. Nor does Newtonian mechanics predict the observed deflection of light by massive objects (gravitational lens effect.) Enough on Newtonian mechanics for now.
What about Southern Hemisphere Sea Ice Data?
Fig. 2 Southern Hemisphere Sea Ice Area, Seasonal Variation by Year.
This data from agencies NOAA, NSIDC, and University of Bremen. Original data plot from this site: http://arctic.atmos.uiuc.edu/cryosphere/arctic.sea.ice.interactive.html
Briefly, the extensive data in this composite plot shows seasonal variations in southern hemisphere sea ice extent over 34 years from 1979 through 2013. We observe that record high sea ice extent occurred in 2013 yellow curve, and in 2012, red curve.
Of course, climate change is inevitable, but a 34 year data set is a rather small time period compared to the historic time scales of climate changes which occur on time scales of multi-decades, centuries, and millennia. The data presented in Fig. 2 indicate a rather stable climate over the timescales available so far.
Back to climate models.
Climate models used and misused to predict the future are more accurately termed climate forecasting models. Climate forecasting models get no credit for predicting the past.
We can summarize the predictive performance of climate models by box scores below. The best score achievable by a model that demonstrates significant long term agreement with all new climate data sets would be 1.
How are climate models doing?
Here are the box scores:
BOX SCORE 2013
Extravagant claims of accuracy
Inevitable failure of extrapolation
BOX SCORE 2013
IPCC climate model extrapolation accuracy: 0
Synthetic Information prediction of failure: 1
We will add further examples of failure of climate models on extrapolation as time permits.
What about modeling of much simpler systems?
How's that going?We observe that the US National Ignition Facility (NIF) at Lawrence Livermore Labs in California was extensively modeled by fluid codes that are much more sophisticated and accurate at what they do, than are climate models.
What is NIF?
The NIF generates 192 powerful laser beams of UV light that are used to compress fusion targets aka pellets and heat fusion fuel in the target so as to achieve ignition of nuclear fusion reactions. Fluid modeling codes were used to predict the performance of the NIF.
Those state of the art fluid code models were used to validate the design of the NIF, and predicted that the NIF should achieve ignition of a nuclear fusion reaction in carefully designed targets containing fusion fuel.
The NIF was constructed and put into operation a few years ago. After a year long campaign of experiments in 2012 scientists found that NIF targets did not reach ignition conditions in the lab.
The NIF is a fantastic scientific instrument for studying matter at extreme conditions of pressure and temperature. The scientific value of NIF is extraordinary.
We might say that the most important science from NIF so far was the invalidation of the fluid models. Such knowledge gained by rigorous testing of model predictions by means of precise experiments is central to scientific progress.
UPDATE ON NIF: FEB 12, 2014
Lawrence Livermore NIF team announces achievement of fusion energy gain greater than unity using a redesigned laser pulse strategy.
Congratulations to the NIF team.
"Energy gain greater than unity" means fusion energy generated by the laser driven implosion of the target is greater than the energy deposited in the fusion target by the laser beam pulse. An important milestone has been reached in the quest for fusion ignition.
However, energy gain of unity is not fusion ignition. Ignition would require a much larger energy gain, and that has not been achieved yet by NIF.
Roughly speaking, ignition means that after an initiating laser pulse, the fusion fuel would continue to generate fusion energy (to burn) in a self-sustaining nuclear fusion reaction. Like a miniature sun, the fusion target would continue to shine on its own, until its fuel is exhausted.
The Earth's climate system is a much more complex and challenging problem for fluid code modelers than is the NIF. It will take many decades of advances in the state-of-the-art of climate modeling to approach the accuracy of the Livermore fluid codes. Until a long track record of correct predictions of the actual future behavior of the Earth's climate is demonstrated by climate models, the models cannot be considered ready for prime-time.
What's the Greenhouse Effect?
Attributed to early writings of astronomer Carl Sagan, the Greenhouse effect refers to absorption of "heat radiation" by atmospheric gases of a planet e.g Venus.
What is Heat Radiation?
Heat radiation is electromagnetic radiation (light waves) emitted by any warm substance as a consequence of thermodynamics. It's the reason hot steel glows in red, orange, yellow colors when heated in a blacksmith's forge, and the reason the Sun is yellow in color. Heat radiation can be viewed as a gas of photons having a temperature comparable to the temperature of the surrounding material surfaces.
The wavelength spectrum of heat radiation in a vacuum enclosure is well described by the Planck radiation theory which predicts a continuous wavelength spectrum given mathematically by the Planck Radiation Formula.
If the heat radiation is emitted into a gaseous material, the gas molecules will interact with the electromagnetic waves and may selectively absorb electromagnetic waves of certain wavelengths. A broad wavelength spectrum of electromagnetic radiation passing through a gas will therefore be attenuated in certain wavelength ranges called "absorption bands."
....to be continued. (2/18/2014)
For example, if there was a steady (monotonic) decrease in mid-September sea ice area from 2005 through 2013, (as predicted by global warming climate models) then the lowest mid-September coverage would have occurred in 2013. Every previous year would have higher mid-September sea ice area, and 2005 would have the highest mid-September sea ice area out of the set of all years in the sample.
Now we can analyze the data in Fig. 1, and see what we get.
Note: We use the original higher resolution figure on the DMI website to read off sea ice area for each year in mid-September. Then we order years by relative minimum sea ice coverage in mid-September. That is, the lowest mid-September minimum occurred in 2012, the next lowest in 2007, and so on, for each year 2005 through 2013.
Here's our data collected from Fig. 1:
Lowest mid-September sea ice area occurred in 2012, blue curve; next lowest 2007, light blue curve; next lowest 2008, magenta; next lowest 2011, yellow; next lowest 2010, orange; next lowest 2009, baby blue or cyan; next lowest 2013, black, next lowest 2005, red, and highest sea ice coverage in mid-September occurred in 2006, green curve.
That is, there is no systematic, year-to-year downward or upward trend, in northern hemisphere mid-September sea ice coverage, for the years for which we have mid-September data, 2005 through 2013.
One can see why this is so by examining year-to-year, increases or decreases in mid-September sea ice area in our data set above:
From 2005 to 2006 the ice area increased, from 2006 to 2007 ice area decreased, from 2007 to 2008 ice area increased, from 2008 to 2009 ice area increased, from 2009 to 2010 ice area decreased, from 2010 to 2011 ice area decreased, from 2011 to 2012 ice area decreased, from 2012 to 2013 ice area increased. We have four cases of increasing ice area and four cases of decreasing ice area over the period 2005 through 2013. Hence no trend of increase or decrease in mid-September sea ice area over the period 2005-2013.
Of course, there are many ways to analyze the data in Fig. 1. The above analysis gives a good indication of the existence or non-existence of a trend in a simple, easy to understand way. The above ordering of years by mid-September sea ice coverage area shows no systematic downward or upward trend.