February 28, 2007

Numerical Models, Integrated Circuits and Global Warming Theory

By Jerome J. Schmitt
Global warming theory is a prediction based on complex mathematical models developed to explain the dynamics of the atmosphere. These models must account for a myriad of factors, and the resultant equations are so complex they cannot be solved explicitly or "analytically" but rather their solutions must be approximated "numerically" with computers.  The mathematics of global warming should not be compared with the explicit calculus used, for example, by Edmund Halley to calculate the orbit of his eponymous comet and predict its return 76 years later.

Although based on scientific "first principles", complex numerical models inevitably require simplifications, judgment calls, and correction factors.  These subjective measures may be entirely acceptable so long as the model matches the available data -- acceptable because the model is not intended to be internally consistent with all the laws of physics and chemistry, but rather to serve as an expedient means to anticipate behavior of the system in the future. However, problems can arise when R&D funding mechanisms inevitably "reward" exaggerated and alarming claims for the accuracy and implications of these models. 

Many other scientific fields besides climatology use similar models, based on the same or related laws of nature, to explain and predict what will happen in other complex systems.  Most famously, the US Department of Energy's nuclear labs use supercomputer simulations to help design atomic weapons. Most of this work is secret but we know, of course, that the models are "checked" occasionally with underground test explosions. The experimental method is an essential tool

A much better analogue to climate science is found in the semiconductor industry. Integrated circuits and many other building blocks of modern electronics are manufactured by creating artificial atmospheres or "climates" within which chemical vapor deposition (CVD) forms nanometer-scale thin solid films on silicon wafer surfaces. In CVD, metal vapor precursors entrained in carrier gases are used to deposit metal films on surfaces in a condensation process not unlike formation of dew or frost on a lawn.  In such CVD processes, premature formation of metal particles is unwanted and needs to be controlled and prevented; such particle formation is akin to precipitation of rain drops in the atmosphere

The semiconductor process industry uses numerical models to predict the behavior of gases and vapors in order to deposit these substances on substrates, and thereby manufacture integrated circuits. I am not a climatologist or meteorologist but I have studied fluid mechanics and gasdynamics and have a general understanding of computer models used in process engineering.  Such models are used to analyze industrial processes with which I am familiar.  Indeed the mathematics for such models is generalized.  And industry's experience with numerical process models sheds light on their strengths and limitations.

Andrew Grove PhD is a giant in the history of semiconductors. A founder of Intel, Grove famously presided as CEO over its enormous growth during the 1980s and 1990s. Few realize that his academic training is as a Chemical Engineer, not an Electrical Engineer.  Chemical Engineering is at the heart of what Intel and other semiconductor manufacturers accomplish. 

Process Models: Vapor deposition

Let's consider how these process engineering mathematical models are actually used in industry.  Intel and its competitors (as well as their key suppliers) employ many chemical engineers who are familiar with such process models, some of whom specialize solely in mathematical modeling.   Often a new technical challenge will emerge in which a process must be changed (such as for scale-up to accommodate larger silicon wafers) or adjusted to accommodate a new material composition. 

Almost all semiconductor manufacturing processes occur in closed vessels.  This permits the engineers to precisely control the input chemicals (gases) and the pressure, temperature, etc. with high degree of precision and reliability.  Closed systems are also much easier to model as compared to systems open to the atmosphere (that should tell us something already).  Computer models are used to inform the engineering team as the design the shape, temperature ramp, flow rates, etc, etc, (i.e. the thermodynamics) of the new reactor.

Nonetheless, despite the fact that 1) the chemical reactions are highly studied, 2) there exists extensive experience with similar reactors, much of it recorded in the open literature, 3) the input gases and materials are of high and known purity, and 4) the process is controlled with incredible precision, the predictions of the models are often wrong, requiring that the reactor be adjusted empirically to produce the desired product with quality and reliability.

The fact that these artificial "climates" are closed systems far simpler than the global climate, have the advantage of the experimental method, and are subject to precise controls, and yet are frequently wrong, should lend some humility to those who make grand predictions about the future of the earth's atmosphere.

So serious are the problems, sometimes, that it is not unheard of for an experimental reactor to be scrapped entirely in favor of starting from scratch in designing the process and equipment. Often a design adjustment predicted to improve performance actually does the opposite.  This does not mean that process models are useless, for they undergird the engineer's understanding of what is happening in the process and help him or her make adjustments to fix the problem.  But it means that they cannot be relied upon by themselves to predict results. These new adjustments and related information are then used to improve the models for future use in a step by step process tested time and again against experimental reality. 

In actuality, the semiconductor industry is well familiar with the limits of process modeling and would never make a decision to purchase equipment or adjust their manufacturing processes based on predictions derived from models alone.  They would rightly expect extensive experimental data to support such a decision in order to assure the ability to reliably and economically manufacture high quality materials and devices. 

Climate Models

As with all fluid mechanics models, the flow field of a climate model (i.e. the entire atmosphere) is divided into three-dimensional grids of small volume elements designated by latitude, longitude and altitude. Each volume element of the grid is then characterized with parameters such as pressure, temperature, wind velocity, etc., and equations that relate these factors.  Air and energy that leave one volume element enters the adjacent one.  When summed across all volume elements, the model keeps track of the flows of air and energy in the entire atmosphere.  Many factors must be accounted (see below).  Boundary conditions must be set: in this case, the boundary of the atmosphere is land or ocean surface on the bottom, and some boundary in space on the top; these yield rules (e.g. air cannot flow into the surface of the earth).  Then, Initial Conditions must be set: this means that the grid's equations are "populated" with the known values of the parameters characterizing the atmosphere such as pressure, temperature, and humidity profiles measured today. 

Finally, the computer calculation can commence:  A unit of time (a second, minute, day) is assumed to pass and the computer calculates the next "state" of the model based on the initial conditions, the boundary conditions and the other equations of the model.  This process is repeated again and again, with the new state being the initial condition for calculating the subsequent state, until e.g. 100 years has passed.

Errors can accumulate rapidly.  Let's list some of the factors that must be included (by no means an exhaustive list):
Solar flux
Gravity, Pressure
Earth's rotation
Surface temperature
Currents in the Ocean (e.g., Gulf Stream)
Greenhouse gases
CO2 dissolved in the oceans
Polar ice caps
Infrared radiation
Cosmic rays (ionizing radiation)
Earth's magnetic field
Cloud formation
Reflection from clouds
Reflection from snow
Soot formation
Trace compounds
And many, many others
Even if mathematics could be developed to accurately model each of these factors, the combined model would be infinitely complex requiring some simplifications.  Simplifications in turn amount to judgment calls by the modeler.  Can we ignore the effects of trace compounds?  Well, we were told that trace amounts of chlorofluoro compounds had profound effects on the ozone layer, necessitating the banning of their use in refrigerators and as aerosol spray propellants.  Can we ignore cosmic rays?  Well, they cause ions (electrically charged molecules) which affect the ozone layer and also catalyze formation of rain-drops and soot particles. 

As with all models, it is perilous to ignore factors in the absence of complete experimental data which might have otherwise have significant effect.

Perhaps most critically, the role of precipitation in climate seems to be understated in the numerical global climate models. Roy W. Spencer, principal research scientist at the Global Hydrology and Climate Center of the National Space Science and Technology Center in Huntsville, AL, writes that the role of precipitation is not fully accounted for in global warming models. In my view, that's like an economist admitting his theory of the money supply doesn't fully account for the role of the Federal Reserve.
Unless we know how the greenhouse-limiting properties of precipitation systems change with warming, we don't know how much of our current warmth is due to mankind, and we can't estimate how much future warming there will be, either. To solve the global-warming puzzle, we first need to learn much more about the precipitation-system puzzle.

What little evidence we now have suggests that precipitation systems act as a natural thermostat to reduce warming.
Approximating the experimental method

While mankind cannot experiment on the global climate, these models can be used retroactively to see how well they "model" the past.  The UN's 2001 Climate Change report distorted the historical record by eliminating the Medieval Warm Period in the famous "Hockey Stick Curve" which, by many accounts, unreasonably accentuated temperature rise in the 20th century.  Such distortion of the historical data undercuts the credibility of the models themselves, since this is the only "experimental data" available for testing the fidelity of the models to the actual climate.

Why on earth would climate scientists "massage the data" to produce doomsday predictions? The answer requires looking at the rewards available to these researchers.

Catastrophe and careers

Vannevar Bush's seminal 1944 policy paper unleashed the Federal government's unprecedented post-war investment in R&D in the hard sciences and engineering. Science was seen as the way to avoid (or at least win) another catastrophic war. 

The golden era of federal funding resulted in unprecedented employment opportunities for hard science Ph.D.s.  Fresh graduates could easily find tenure track employment at universities expanding their hard sciences program. The enormous dividends from this investment make up our modern technological world.  However, the munificence of the federal funding caused a certain, shall we say, insouciance about resources: "Why use lead when gold will do?" became an informal motto at Lawrence Livermore National Lab.

Inevitably, the growth in congressional funding tapered off and in the late 1980s the competition for R&D sponsorship began to tighten.  Fresh Ph.D.s often had to look to the private sector for employment (heaven forefend!).  Grant writers were required to start highlighting the potential "practical payoffs" of their proposed work.  Since there was little need for better atomic weapons in the post-cold war era, High Energy Physics lost its central status in the funding universe.  Many mathematical physicists became refugees to allied fields (some of them even became "quants" on Wall Street). But others found employment elsewhere, including in climate science.

In this competitive environment, one can imagine climate modelers justifying their work by citing the possibility of global change, the further study of which requires, of course, "more research".  One can further imagine that in the inchoate communication between university researcher, funding agency, congressional staffer and congressmen that "possibility" eventually became "probability" and then "probability" morphed into "certainty" of global warming, especially if there was potential for political advantage. 

This has resulted in an inadvertent funding-feedback mechanism that now resonates in largely unjustified alarm and also seeks to quash scientific dissidents who indirectly threaten to throttle the funding spigots.

The practical experience of numerical modeling in allied fields such as semiconductor process modeling should cause us to question the claimed accuracy for Global Climate Models.  The UN's distortion of historical climate data should further undermine our faith in climate models because such models can only be "tested" against accurate historical data. 

In my view, we should adopt the private sector's practice of placing extremely limited reliance on numerical models for major investment decisions in the absence of confirming test data, that is, climate data which can be easily collected just by waiting. 

Jerome Schmitt is president of NanoEngineering Corporation, and has worked in the process equipment and instrument engineering industries for nearly 25 years.