From Dr. Roy Spencer’s Global Warming Blog
by Roy W. Spencer, Ph. D.
This is the third in my (seemingly never-ending) series on measuring the impact of the Urban Heat Islands (UHI) on land surface temperature trends.
IN my part And Part II I have highlighted the Landsat-based “construction” structural dataset as a proxy for urbanization, which I am sure we will continue to examine as part of the Department of Energy funding. to test (most) satellite-based methods and datasets to test climate models and their predictions about global warming.
Much of the early research on the UHI effect (eg: TR Oke, 1973 and later) with respect to the warming of the total population of towns and cities. Since population data sets span much further back in time than the satellite period, they can provide information about the UHI effect that goes back to pre-1900. Over the past few weeks I have turned to using Using Landsat satellite-based settlement diagnoses, construct structures as a proxy for urbanization, for population density. Along the way, I had to investigate problems related to low correlation and linear regression (namely, regression dilution). I decided not to mention that here because it’s a bit too technical.
The more I dig into this project, the more I learn.
Urbanization effect from 1880 to 2015
I have a lot of results to present, but I thought I would introduce only one plot of interest. Using tens (in the early years) to hundreds of thousands of temperature difference pairs and PD differences at 2 stations, I sorted them from smallest 2-station average PD to largest. Then I do the regression in separate PD periods (12 to 19 of them) to get the change in temperature by population density (dT/dPD). In fact, these coefficients expose a non-linear function that relates PD to the warming effect of UHI. The data shown below is from June for a 20-year period from 1880 to 2015, in the 20N to 80N latitude range.
Then, by summing those regression coefficients (integrate them, in analytical terms) from 0 PD to the maximum mean PD value of the 2 stations, I construct curves of the PD effect versus with UHI. I have looked at quite a few published UHI articles and I cannot find a similar approach to the UHI problem.
I must admit, the results in Figure 1 are not what I expected. They show that the total UHI effect was stronger at the end of the 19th century and has weakened somewhat since then. (Remember, since these results are based on differences between the 2 stations, these are spatial relationships, i.e. over the period 1880, 1890, 1900, there was a larger temperature difference between regions. rural and densely populated areas compared with the following decades.)
I have no ready explanation for this, and ideas are welcome.
If the results are reversed, I would guess it’s because of the larger error in the original population estimate, since the error in the independent variable (PD) lowers the regression slope (dT/dPD) below the relationship “correct” (regression dilution). But the opposite is happening. And, it can’t be because the number of stations is much lower in the early period as that only leads to noise in the regression coefficients and not the system bias.
A few thoughts
From reading the literature, I think this is a rather novel approach that avoids a common problem: the usual division of stations into ‘rural’ and ‘urban’ categories. Because the curves in Figure 1 are non-linear, a station near the countryside will experience more heating due to a given population increase than a very urban area. Therefore, previous investigations showing little difference in temperature trends between urban and rural areas do not really prove anything.. My method avoids that problem by constructing curves starting from zero population density (really rural conditions).
Ultimately, all of this will lead to estimates of how much land warming (such as since 1880) is spurious due to the Urban Heat Island effect. As I mentioned before, I don’t believe it will be that big. But it needs to be recorded.
Related
