Urban warming impacts are usually evaluated by comparing temperature gradients in urban areas with temperature gradients in adjacent rural areas. This post approaches the question from a different perspective. Instead of looking at urban warming per se it looks at the urban heat islands that cause it. How common are they? How variable are they? How large are the areas they cover? What are their amplitudes? Knowing the answers to these questions could help improve our understanding of how much of the increase in global surface air temperature over the period of instrumental record might have been caused by urban warming. Alternatively it might confuse us even further, but that’s a form of progress too.
Unfortunately very little quantitative information is available on urban heat islands, or at least it wasn’t some years ago when I did the work reported here. The available data consisted – and as far as I can see still consist – of temperature profiles across idealized metro areas, infrared satellite images, maps showing metro area temperatures on a hot summer afternoon or on a cold winter morning, temperature profiles measured by car thermometers in cross-city vehicle trips and localized studies such as the temperature difference between parks and office blocks in Singapore. None of this was of any use in quantifying heat island extents and amplitudes, which requires long-term temperature data from weather stations both inside and outside the urban area.
So I developed a data set from scratch. It quantifies at variable levels of precision and detail the UHI signatures of 43 different metro areas (28 in the USA, 15 in other countries) based on annual mean temperature data from 930 stations, and in this post I summarize the results.
A couple of disclaimers before proceeding. First, I can no longer find a link to the NOAA site I downloaded the US data from (the non-US data were from GISS GHCNv2). You will just have to believe me when I say I didn’t make them up. Second, the temperatures I show are averages of annual means mostly centered around 1975-1980, so they will tend to understate the size of the present-day urban heat island.
And a technical note: All station temperatures are adjusted for zero correlation with latitude and elevation in each urban area.
Now to the results. The end product of the work was a portfolio of maps showing temperature distributions in and around the 43 urban areas, and Figure 1 shows one of them – Los Angeles, California, metro area population 12.9 million and according to my data the largest urban heat island in the world:
Figure 1: Los Angeles Urban Heat Island
Temperatures in the LA metro area average 17.9C and temperatures outside average 16.1C, a difference of 1.8C. Multiplied by the 2,300 square mile extent of the LA metro area this gives a UHI with a size of 4,140 degrees-miles-squared. This number is a crude measure of the total amount of heat generated by the UHI.
Coming in second is New York, metro area population 19.4 million (Figure 2). The New York UHI is concentrated around the downtown “concrete canyon” urban area, but if the entire metro area (not outlined) is considered there is a temperature difference of 1.0C – mean temperature 11.2C inside versus 10.2C outside – over an area of 2,250 square miles, giving 2,250 degrees-miles-squared. (Why is the Los Angeles UHI almost twice as large? Probably because LA is ringed by mountains that confine heat within the LA basin.)
Figure 2: New York urban heat island
Third comes Chicago, metro area population 9.8 million, with 1,200 degrees-miles-squared, and after that Miami with 900 and so on down the list.
What about urban areas outside the US? Well, there weren’t many where GHCN station density was sufficient to allow me to draw a meaningful UHI map. Because of higher population density urban areas outside the US also tend to be smaller and the size of the UHI correspondingly lower. An example is Osaka-Kobe-Kyoto, metro area population 18.7 million, shown in Figure 3. Mean temperatures inside and outside the metro area are 1.1C different, about the same as New York, but because the metro area is only a third the size (700 vs. 2,250 sq miles) the UHI is only about a third the size too.
Figure 3: Osaka-Kobe-Kyoto urban heat island
At the other end of the list are a number of urban areas that for whatever reason show no detectable UHI, including Atlanta, Cape Town and Manila. Probably the best example is the Washington-Baltimore metro area, population 8.4 million, shown in Figure 4. Apart from the single 14.2C value in downtown Baltimore it has no detectable UHI. Temperatures inside and outside the metro area are effectively the same.
Figure 4: Washington-Baltimore urban heat island, or lack thereof
I can’t discuss the results for each urban area individually so I’ve summarized them in Figure 5, which plots temperature anomalies relative to the mean temperature outside the urban area for the 930 stations in the 43 areas against distance from the urban center. Distances are normalized relative to an “average urban area” which, if it existed, would be thirty miles in diameter and have a population of 6.6 million. (There is no really satisfactory way of combining the data from all 43 areas, but plotting them without normalizing distance mixes stations inside larger urban areas with stations outside smaller urban areas and leads to distorted results.)
Figure 5: Temperature anomalies relative to mean local background temperature against normalized distance from the urban center, all data.
There’s a lot of scatter, but except for the blip around 55 miles averaging the data gives a respectably smooth curve (red line). This curve defines a UHI with a maximum amplitude of about 0.9C which is detectable out to a distance of around 30 miles from the urban center, or up to 15 miles outside the city limits of our “average urban area”.
I did, however, investigate one other interesting aspect of urban heat islands that I will briefly discuss before concluding, and that’s how much of the UHI signature is caused by the presence of an urban area and how much by conditions in the vicinity of the station. I evaluated this by locating the individual stations used to construct Figure 5 as closely as I could on Google Earth and segregating them into three categories:
- Urban – city centers, concrete canyons, heavy industry etc.
- Suburban – houses, schools, offices, shopping centers, roads, trees, gardens, parks etc.
- Rural – open country, maybe a few scattered buildings and roads.
Then I replotted the Figure 5 temperature vs. distance graph with the data segregated into these categories and came up with the results shown in Figure 6:
Figure 6: Temperature anomalies relative to mean local background temperature against normalized distance from the urban center, stations segregated into urban, suburban and rural categories.
The individual urban, suburban and rural points show considerable scatter but their averages show clear systematic differences. The average urban station runs about a degree warmer than the average suburban station and except for the excursion around 37 miles the average suburban station runs consistently about 0.2C warmer than the average rural station regardless of location relative to the urban area. These results suggest that temperatures in and around an urban area are controlled dominantly by conditions around the point of reading, not by distance from the urban center, or in other words that an urban heat island isn’t really an island at all but an archipelago of urban and suburban heat islets. It looks like a single heat island simply because the proportion of suburban and urban stations increases as we go inwards.