What is the Effect of Urban Heat Islands on Global Warming Trends?

Baltimore's Inner Harbor

It is well-documented that cities are significantly warmer than rural areas. This effect is commonly described as an urban heat island (UHI). Some argue that some (or all) of the increase in global temperatures can be explained by the effect of the UHI effect on global temperatures. It makes some intuitive sense that this might be the case, but let's consider some important lines of evidence that suggest otherwise.

1. While urban areas are warmer than rural areas, urban areas are warming at about the same rate as rural areas, so temperature anomalies are not significantly affected by urban heat islands.[1]
2. There is a well-documented bias that can be added to the temperature record from urbanization. As a rural area becomes more urban, it will warm at a faster rate than cities and rural areas. An area that is rural in 1950 but urban in 2020 will become progressively hotter due to urbanization, and not changing climatic conditions. However, numerous studies have shown that this potential source for bias in temperature trends is corrected by bias correction.[2][3] One interesting study evaluating the old USHCN dataset concluded that UHI is insufficient to contaminate the temperature record: “If ∼30% of the highest population stations were removed from the analysis, no statistically significant urban heat island was detected. The implications of this work on U.S. climate change analyses is that, if the highest population stations are avoided (populations above 30 000 within 6 km), the analysis should not be expected to be contaminated by UHIs. However, comparison between U.S. Historical Climatology Network (HCN) time series from the full dataset and a subset excluding the high population sites indicated that the UHI contamination from the high population stations accounted for very little of the recent warming.”[4]
3. Cities take up a small fraction of the globe's surface, so whatever bias may remain in the data after homogenization must be small. Most of the planet's surface is ocean, and most of the land's surface is natural land and rural.
4. In the US, NOAA established the USCRN dataset in 2005, a network with 114 ideally-sighted, all rural stations and no homogenization). In 2014, NOAA replaced USHCN with nClimDiv, which is a network with 10,000 stations and homogenization. The two datasets are virtually indistinguishable from each other during the years when they overlap (2005-2024). This confirms that bias correction is working, successfully removing urbanization bias from the data.[5] In the US, rural, urban, and rapidly developing areas are all generally free from bias from urban heat islands. 
5. Wickham et al 2013 evaluated station data in the Berkeley Earth project globally, and compared the warming rates of all stations vs very rural stations. The study concluded: “Time series of the Earth’s average land temperature are estimated using the Berkeley Earth methodology applied to the full dataset and the rural subset; the difference of these is consistent with no urban heating effect over the period 1950 to 2010, with a slope of -0.10 ± 0.24/100yr (95% confidence).”[6] If anything, rural stations are warming more rapidly than all stations.

Very Rural Stations Warm as Rapidly as All Stations in Wickham et al 2013

If any significant fraction of current warming was attributable to urbanization, the rate at which the most rural stations are warming would be slower than the entire network. But in point of fact, they are warming at virtually the same rate, so it cannot be that urbanization is causing any spurious warming trends. If you doubt the arguments above after evaluating the cited resources and studies, there's another way for you to test this for yourself. There's an online tool you can use to generate graphs from GHCN land station data, and you can select all land stations and two designated rural subsets of stations. The tool uses data from 1900 to 2014, which should be more than enough data to see if global land trends are warming more rapidly than rural areas.

The above graph shows GHCN land station data along side two forms of rural data: one with 2167 station designated rural locations and one with 1506 rural stations identified by areas by the amount of light seen at night. As you can see above, no matter which group of thermometers you use, GMST trends are essentially identical. And it turns out you don't even need all rural stations to approximate the global land trend accurately.

2025 Update: I downloaded two rural time series from the above tool, one with 122 stations and the other with 130 stations, and both were all rural stations detected by two different means. I then plotted these with NASA GISSTEMP v4 global land temperature anomalies. While the rural time series have a bit more variability (as expected with fewer stations), the overall trends are nearly identical to the global land trend. So you can get a good estimate of the global land mean change in temperatures using just ~125 randomly distributed rural land stations.

References:

[1] Jones, P. D., Lister, D. H., and Li, Q. (2008), Urbanization effects in large‐scale temperature records, with an emphasis on China, J. Geophys. Res., 113, D16122, doi:10.1029/2008JD009916.
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2008JD009916

[2] Menne, M.J. and C. N. Williams Jr. “Homogenization of Temperature Series via Pairwise Comparisons.” Journal of Climate 22 (2009): 1700-1717. http://dx.doi.org/10.1175/2008JCLI2263.1

[3] Parker, D. E., 2006: A demonstration that large-scale warming is not urban. J. Climate, 19, 2882–2895.
https://journals.ametsoc.org/doi/10.1175/JCLI3730.1

[4] Peterson and Owen, “Urban Heat Island Assessment: Metadata Are Important” Journal of Climate 18 (2005): 2637-2646.
https://journals.ametsoc.org/doi/pdf/10.1175/JCLI3431.1

[5] https://www.ncdc.noaa.gov/news/how-ncdc%E2%80%99s-datasets-compare-each-other

[6] Wickham C, Rohde R, Muller RA, Wurtele J, Curry J, et al. (2013) Influence of Urban Heating on the Global Temperature Land Average using Rural Sites Identified from MODIS Classifications. Geoinfor Geostat: An Overview 1:2. doi:10.4172/2327-4581.1000104
https://berkeleyearth.org/wp-content/uploads/2022/12/UHI-GIGS-1-104.pdf