The website RealClimate had a very informative set of questions from Tom Cole and answers from Gavin Schmidt . RealClimate provides a valuable service by providing a set of issues in this Q&A format that we can answer. I provide my perspective on the questions below, in order to add to this discussion.
1. What schemes are you using for solving the partial differential equations? Are they free of numerical errors?
No model is free of numerical errors.
Climate models require the accurate simulation of the ocean, atmosphere, land, and continental ice. Physical, chemical, and biological processes must be included. In the atmospheric and ocean components of these models only the pressure gradient force and advection are represented in terms of fundamental concepts. This part of the models is referred to as the “dynamic core.” All other processes in these models are parameterized (e.g., turbulence, cloud and precipitation, short- and long-wave radiative fluxes).
The dynamical core of the models has been represented with finite difference and spectral methods; the latter of which is typical for global models, while regional climate models generally have applied finite differencing. For spatial scales less then 4 grid increments (or its equivalence in a spectral model), there is always serious numerical error (either in terms of preservation of amplitude and/or phase). For finite difference models, this is discussed in detail in Chapter 10 âMethods of Solutionâ? of Pielke, R.A., Sr., 2002: Mesoscale meteorological modeling. 2nd Edition, Academic Press, San Diego, CA, 676 pp.
This inability of models to skillfully simulate the smallest features within the grid structure is why the term “resolution” should be reserved to refer to spatial scales of at least 4 grid increments in each direction. This limitation applies to both finite difference and spectral models (see Pielke, R.A., 1991: A recommended specific definition of “resolution”, Bull. Amer. Meteor. Soc., 12, 1914; Pielke Sr., R.A., 2001: Further comments on “The differentiation between grid spacing and resolution and their application to numerical modeling”. Bull. Amer. Meteor. Soc., 82, 699; and Laprise, R., 1992: The resolution of global spectral models. Bull. Amer. Meteor. Soc., 9, 1453-1454.
Parameterizations that are used in the models have been vertical (i.e., one-dimensional) column or box models, and always include adjustable, tunable coefficients and functions. They are engineering codes which are calibrated based on observations, sometime in conjunction with higher resolution models, usually from what are often referred to as “golden days.” Golden days are selected with ideal conditions in order to best fit the theoretical framework of the parameterization. Since parameterizations are applied in the climate models for situations in which the parameterizations were not calibrated, there certainly are errors but of an unknown magnitude.
A powerpoint presentation that overviewed these issues is available at (Pielke, R.A., Sr., 2004: The Limitations of Models and Observations. COMET COMAP Symposium 04-1 on Planetary Boundary Layer Processes, Boulder, Colorado, June 21-25, 2004.).
2. Have you made tests to determine if the model results depend on resolution? In other words, have you increased the detail sufficiently so that the results are no longer dependent upon the size of an individual grid box?
Model results are always dependent on the grid increments used. It is unreasonable to expect the one-dimensional column and box parameterizations to accurately represent real-world three-dimensional features that are spatially smaller than can be resolved by the model grid increments.
That resolution matters is shown quantitatively in the paper Castro, C.L., R.A. Pielke Sr., and G. Leoncini, 2005: Dynamical downscaling: Assessment of value restored and added using the Regional Atmospheric Modeling System (RAMS). J. Geophys. Res. – Atmospheres, 110, No. D5, D05108, doi:10.1029/2004JD004721. In that paper, Table 5 presents a suite of regional climate experiments for horizontal grid intervals with 50 km, 100 km and 200 km spacing. The degradation in model skill as the horizontal increment is increased is shown.
Another issue, that the global climate models have not adequately addressed, is how well do they perform when initialized in a numerical weather prediction mode in terms of such atmospheric features as extratropical and tropical cyclogenesis? This test is a necessary condition of the accuracy of both the dynamics and parameterizations within the model. Since some global climate models have a parentage from numerical weather prediction code, this would be straightforward for them to evaluate with the code as adapted for long-term climate simulations. Clearly, a global model that is superior to others when it is run as a weather forecast, with observed initial conditions, will be a superior climate model as this means its dynamics and parameterizations are more accurate. Such comparison experiments starting from initial conditions have not been performed and documented in the literature to my knowledge. This approach would be an extension of the Atmospheric Model Intercomparison Project (AMIP) comparisons (i.e., as discussed, for example in Research Activities in Atmospheric and Oceanic Modeling, J. Cote, Ed.).
3. What are the dominant external forcing functions?
Figure 1-2 in the 2005 National Academy report defines natural forcings as from the Sun, due to the Earthâs orbital characteristics, and from volcanoes. Natural as defined here is meant to include forcings which reside external from the climate system. With this definition, the human-climate forcings are not external forcings.
4. What are the sources of intrinsic variability?
We do not know all of the reasons for the intrinsic (internal) variability of the climate system. Gavin has clearly identified some of them. However, the paper by Rial, J., R.A. Pielke Sr., M. Beniston, M. Claussen, J. Canadell, P. Cox, H. Held, N. de Noblet-Ducoudre, R. Prinn, J. Reynolds, and J.D. Salas, 2004: Nonlinearities, feedbacks and critical thresholds within the Earth’s climate system. Climatic Change, 65, 11-38, illustrates observed examples on a variety of time scale of sudden and rapid transitions of climate which we do not adequately understand. One major conclusion, however, is that intrinsic, complex variability results from interfacial nonlinear interactions between the components.
5. How do errors in estimating the forcing functions, or in simulating the internal variability impact the results?
I agree with Gavin that this is a good question. However, until we include all of the first-order climate forcings and feedbacks, as well as successfully model sudden climate transitions, we have large remaining errors of an unknown magnitude. We also have to show prediction skill in the quasi-linear global and regional long term trends of important climate metrics (regional precipitation, regional layer-averaged tropospheric temperatures, etc). Whether or not we agree the models have shown skill in reproducing global temperature averages or not, they certainly have not demonstrated regional skill for the spectrum of important climate metrics.
These first-order climate forcings are identified in the National Research Council report “Radiative Forcing of Climate Change: Expanding the Concept and Addressing Uncertainties”, while the first-order climate feedbacks are identified in “Understanding Climate Change Feedbacks”. Sudden climate transitions are discussed in the National Research Council report “Abrupt Climate Change: Inevitable Surprises” .
6. The minimum amount of observed data that you have to reproduce in order to gain some confidence in your model is that you have to reproduce periods of time when temperatures are increasing and when they are decreasing. Have you queried the model as to what the dominant mechanism(s) is/are that caused the cooling? If so, is/are the mechanism(s) plausible? Can they be verified independently?
Gavin stated “This isn’t much of a test. The models are pretty stable in the absence of forcing changes (although there is some centennial variability as noted above, related mostly to ocean circulation/sea ice interactions).”
As illustrated in the National Research Council reports and in the Rial et al. paper, however, the observations show that the climate system is not “pretty stable” even without clear changes in the external forcing, If the models are unable to skillfully simulate and explain abrupt regional climate change, they are of limited use in describing our real risk to human-caused and natural climate change and variability. We need to move beyond “linear climate change” thinking.
Also, we need to move beyond global mean surface temperature (and even global mean tropospheric temperature) as the primary climate change metrics. This was a clear conclusion of the 2005 National Research Council report . We need to focus on climate metrics such as drought, growing season and floods, for example, which are climate effects that directly impact society and the environment.
7. Have you tested the model against simplified analytical solutions? Are you able to accurately reproduce analytical results?
Gavinâs answer is correct. We need to use observations to test the models. This is one reason that there is considerable interest in using global and regional climate models to simulate prehistoric and historic climates.
8. How do you address the issue that models cannot be used to predict the future? In other words, models can only predict what might happen under a given set of conditions, not what will happen in the future.
The IPCC and US National Assessment results have been interpreted as predictions. To use the word ‘projection’ to indicate they are different than ‘prediction’ is a nuance that is lost by almost everyone. Indeed the Websterâs New World Dictionary (1988 edition) has one definition of a projection as “a prediction or advance estimate based on known data or observations; extrapolations.” A projection is a prediction! See also my July 15 and 22, 2005 weblogs entitled What Are Climate Models? What Do They Do? and Are Multi-decadal Climate Forecasts Skillful? , and my 2002 Climatic Change essay Overlooked issues in the U.S. National Climate and IPCC assessments on this subject.
We do have a serious problem in climate science in that the same individuals who perform the research are completing the climate assessment reports. This is equivalent to the authors of a research paper, and their close collaborators, review their submitted paper! When I served as Chief Editor of the Monthly Weather Review and Co-Chief Editor of the Journal of Atmospheric Science, this type of procedure was never was permitted. It should certainly not be allowed for the CCSP and IPCC reports, and, to the extent it is, those reports should be interpreted as advocacy documents and not a balanced review of climate science (see, for example my October 4 2005 weblog entitled “Overlooked Issues in Prior IPCC Reports and the Current IPCC Report Process: Is There a Change From the Past?.”
10. I have been working on the same code for over 27 years, and I can guarantee that it is not bug free. A debuggers job is never done. How long has your code been in development?
The more serious error in the models is their incomplete representation of the climate system including the accurate representation of all first-order climate forcings and feedbacks. We also need to know the sensitivity of the model results due to the uncertainly in the parameterizations and from the spatial resolution used. Coding bugs, while as anyone who has written code realizes, never completely disappear as the model is applied to new situations, is not a major problem with climate model simulations that I am aware of.
As my final comment, I want to add to Gavinâs closing remarks, reproduced below
“On a final note, an implicit background to these kinds of questions is often the perception that scientific concern about global warming is wholly based on these (imperfect) models. This is not the case. Theoretical physics and observed data provide plenty of evidence for the effect of greenhouse gases on climate. The models are used to fill out the details and to make robust quantiative projections, but they are not fundamental to the case for anthropogenic warming. They are great tools for looking at these problems though.”
Models are a powerful tool to better understand the climate system and to assess the sensitivity of the climate system to human and natural climate forcings. They have shown us that the radiative effect of the of addition of greenhouse gases is a first-order climate forcing that alters our climate.
However, where I and others disagree with Gavin is the statement that “The models are used to fill out the details and to make robust quantitative projections⦔. What “details” and what demonstration of “robust quantitative projectionsâ??This blanket statement needs to be clarified. Even Mike MacCracken and colleagues for example, have published a paper in Nature in 2004 entitled “Reliable regional climate model not yet on horizon.” The overselling of regional and global models as skillful (robust) projections, unfortunately, rather than as sensitivity simulations, adds to the existing politicalization of climate science and provides justifiable criticism of the assessment reports that are published.
Climate Science: Roger Pielke Sr. 