Costs of Climate Change in Perspective

This is a draft proposal in which to frame our thinking about the climatic impacts of global warming, without getting lost in trivial details, or questioning motives. This builds upon my replication of the thesis of the Stern Review in a graphical form, although in a slightly modified format.

The continual rise in greenhouse gases due to human emissions is predicted to cause a substantial rise in average global temperatures. This in turn is predicted to lead severe disruption of the global climate. Scientists project that the costs (both to humankind and other life forms) will be nothing short of globally catastrophic.

That is

CGW= f {K}                 (1)

The costs of global warming, CGW are a function of the change in the global average surface temperatures K. This is not a linear function, but of increasing costs per unit of temperature rise. That is

CGW= f {Kx} where x>1            (2)

Graphically


The curve is largely unknown, with large variations in the estimate of the slope. Furthermore, the function may be discontinuous as, there may be tipping points, beyond which the costly impacts of warming become magnified many times. Being unknown, the cost curve is an expectation derived from computer models. The equation thus becomes

E(CGW)= f {Kx}                (3)

The cost curve can be considered as having a number of elements the interrelated elements of magnitude M, time t and likelihood L. There are also costs involved in taking actions based on false expectations. Over a time period, costs are normally discounted, and when considering a policy response, a weighting W should be given to the scientific evidence. That is

E(CGW)=f {M,1/t,L,│Pr-E()│,r,W}    (4)

Magnitude M is the both severity and extent of the impacts on humankind or the planet in general.

Time t is highly relevant to the severity of the problem. Rapid changes in conditions are far more costly than gradual changes. Also impacts in the near future are more costly than those in the more distant future due to the shorter time horizon to put in place measures to lessen those costs.

Likelihood L is also relevant to the issue. Discounting a possible cost that is not certain to happen by the expected likelihood of that occurrence enables unlikely, but catastrophic, events to be considered alongside near certain events.

│Pr-E()│ is the difference between the predicted outcome, based on the best analysis of current data at the local level, and the expected outcome, that forms the basis of adaptive responses. It can work two ways. If there is a failure to predict and adapt to changing conditions then there is a cost. If there is adaptation to anticipation future condition that does not emerge, or is less severe than forecast, there is also a cost. │Pr-E()│= 0 when the outturn is exactly as forecast in every case. Given the uncertainty of future outcomes, there will always be costs incurred would be unnecessary with perfect knowledge.

Discount rate r is a device that recognizes that people prioritize according to time horizons. Discounting future costs or revenue enables us to evaluate the discount future alongside the near future.

Finally the Weighting (W) is concerned with the strength of the evidence. How much credence do you give to projections about the future? Here is where value judgements come into play. I believe that we should not completely ignore alarming projections about the future for which there is weak evidence, but neither should we accept such evidence as the only possible future scenario. Consider the following quotation.

There are uncertain truths — even true statements that we may take to be false — but there are no uncertain certainties. Since we can never know anything for sure, it is simply not worth searching for certainty; but it is well worth searching for truth; and we do this chiefly by searching for mistakes, so that we have to correct them.

Popper, Karl. In Search of a Better World. 1984.

Popper was concerned with hypothesis testing, whilst we are concerned here with accurate projections about states well into the future. However, the same principles apply. We should search for the truth, by looking for mistakes and (in the context of projections) inaccurate perceptions as well. However, this is not to be dismissive of uncertainties. If future climate catastrophe is the true future scenario, the evidence, or signal, will be weak amongst historical data where natural climate variability is quite large. This is illustrated in the graphic below.


The precarious nature of climate costs prediction.

Historical data is based upon an area where the signal of future catastrophe is weak.

Projecting on the basis of this signal is prone to large errors.

In light of this, it is necessary to concentrate on positive criticism, with giving due weighting to the evidence.

Looking at individual studies, due weighting might include the following:-

  • Uses verification procedures from other disciplines
  • Similarity of results from using different statistical methods and tests to analyse the data
  • Similarity of results using different data sets
  • Corroborated by other techniques to obtain similar results
  • Consistency of results over time as historical data sets become larger and more accurate
  • Consistency of results as data gathering becomes independent of the scientific theorists
  • Consistency of results as data analysis techniques become more open, and standards developed
  • Focus on projections on the local level (sub-regional) level, for which adaptive responses might be possible

To gain increased confidence in the projections, due weighting might include the following:-

  • Making way-marker predictions that are accurate
  • Lack of way-marker predictions that are contradicted
  • Acknowledgement of, and taking account of, way-marker predictions that are contradicted
  • Major pattern predictions that are generally accurate
  • Increasing precision and accuracy as techniques develop
  • Changing the perceptions of the magnitude and likelihood of future costs based on new data
  • Challenging and removal of conflicts of interest that arise from scientists verifying their own projections

    Kevin Marshall

Leave a comment

4 Comments

  1. Brian H

     /  October 27, 2012

    Very cogent and scholarly. Unspoken, of course, is the fact that Climate Science, as she is done, accommodates few of those criteria. I think, if you were to program that weighted cost curve and apply it to CS, you would get a robust contrast in favour of play-it-as-it-comes versus Precautionary Panic.

    Reply
  1. Stephan Lewandowsky on Hurricane Sandy « ManicBeancounter
  2. Lamar Smith and Implementing effective policy on climate change | ManicBeancounter
  3. 5 The Climate Cost Curve | ManicBeancounter

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 32 other followers

%d bloggers like this: