September 30, 2007

GIGO Adaptation

Berlaymont building morningProjecting Heat-Related Mortality Impacts Under a Changing Climate in the New York City Region -- Knowlton et al., 10.2105/AJPH.2006.102947 American Journal of Public Health analyses the effect of climate change on heat-related premature mortality. They get increases in the 2050s between 47% to 95%, with a mean 70% increase compared with the 1990s, and a decrease of these estimates by 25% due to acclimatization (i.e. being used to it, more use of air conditioning).

What is wrong with this picture?

Leaving aside the uncertaintities of the climate model, the authors make use of current data for the population health response to heat, and use current data from cities with similar climate as the 2050 New York to estimate the effect of acclimatization.

What I'd like to see is a validation of this method by using 1963 US data on mortality and air conditioning use in today's climate. In 1963 domestic air conditioning had just recently become affordable and was still a middle class status symbol, the death rate across the population was 17% higher and the (real, inflation adjusted) GDP per capita was just $15036 compared to $37807 in 2006. Does anybody imagine it would have been even remotely right?

Assuming that the economic trends continue we should have a 250% richer population in 2050, air conditioning would have had the time to become cheaper and more widespread, the demographic structure would have become completely different and in particular, if heat-related mortality is viewed as a problem by the future people, they would also have worked on solving it. While building a model on the real data of today is far better than just making something up, it still has serious problems extrapolating this far. They claim:

Although considerable uncertainty exists in climate forecasts and future health vulnerability, the range of projections we developed suggests that by midcentury, acclimatization may not completely mitigate the effects of climate change in the New York City metropolitan region, which would result in an overall net increase in heat-related premature mortality.

What they have actually shown is that 2007 acclimatization would not mitigate a possible 2050 weather. Which is a bit like arguing that 1963 car safety efforts would be unable to mitigate traffic deaths in 2007 very much.

Both this study and an earlier one seem to make use of weather-based mortality data averaged from 1964-1991. This means that any change over time in this 30 year interval gets removed. it would be far more interesting to extract this change and try to extrapolate it, even if the extrapolation would introduce far more uncertainty in the final numbers. Better a broad confidence interval that tells us something true but uncertain than a narrow interval we can be certain is not true.

While no doubt a changed climate can bring nasty problems, one should be very careful about extrapolating from the present. At the very least one should check for things like Garbage in, garbage out in any model. But running things through a computer and especially invoking climate change seems to excuse anything.

Posted by Anders3 at September 30, 2007 03:37 PM