Our paper (jointly written by John Christy and Danny Braswell) on the calculation of the urban heat island effect (UHI) as a function of population density (PD) is currently in the final stage of review after the 3rd edit, and I’ hope to soon be accepted for publication. So far, I’ve only used Tavg data (average daily maximum and minimum temperature) to develop and test the method, and the paper has only used nearby US summer data (June, July, August), which I’ll talk about. here.
These methods allow us to calculate UHI trends using a global gridded PD data set going back to the 1800s. This UHI trend can then be compared with the temperature trend of the GHCN station. If I do this for all US GHCN stations that have at least 120 years of complete monthly data (June, July, or August) out of 129 potential years during 1895-2023, the plot below shows some interesting results. (I start with the “raw” data so we can examine how homogenization changes the results.)
- The larger the station population, the greater the observed warming trend. This is pretty convincing evidence that the raw GHCN data has a substantial UHI effect that drives the calculated trend (probably no surprise here). Note that the UHI temperature trend is an average of 66% of the raw temperature trend.
- A regression line fit to the data intercepting zero indicates that stations where there is no population growth, on average, there is no warming trend. Although this may lead some to conclude that there was no net warming in the US during 1895-2023, it should be noted that this is raw data, without adjustments for changes in time of observation (TOBS) or changes in the type of instrumentation that may have occurred. bias most or all stations toward a lower temperature trend.
Since most of the rural stations (many of which are experiencing low population growth) are in the western US, and there may be a real difference in temperature trends between the eastern and western US, let’s see how things change if we only examine the eastern US (Ohio to Florida peninsula, east):
This shows the Eastern US has similar features to the US as a whole, with the zero regression line intercept (again) indicating stations with no (average) population growth and no warming trend in the raw GHCN data. But now, amazingly, the average UHI trend is more than 95% of the raw station trend (!) This would represent all the warming reported during 1895-2023 in the eastern US due to urbanization effects… if there is no systematic bias in the raw Tavg data that would cause the trend to be less biased. Also, as will be discussed below, this is the period 1895-2023… the results for more recent decades are slightly different.
Homogenization of GHCN Data Produces Some Strange Effects
Next, let’s see how the adjusted (homogenized) GHCN temperature trend compares to the UHI warming trend. Note that the Pairwise Homogenization Algorithm (PHA) used by NOAA to create the “regulated” GHCN data set (which is the basis of the official temperature statistics coming from the government) identifies and adjusts for time changes at individual stations by comparing temperatures. time series to time series from surrounding stations. If we plot the adjusted data trend along with the raw data trend, the figure below shows some strange changes.
Here’s what the homogenization has done to the raw temperature data:
- Stations with no population growth (which on average show no warming trend) now have a warming trend. I can’t explain this. It is possible that the “urban mixing” artifact of the PHA algorithm discussed by Katata et al. (2023, and references therein) where homogenization does not set urban stations to “look like” rural stations, but tends to narrow the differences between neighboring stations, causing a “bleeding” of urban effects into rural stations.
- Stations with large population growth have reduced the warming trend. This is the desired homogenizing effect.
- There is still a UHI effect on the homogenization trend, but it has been reduced by about 50% compared to the raw trend.. This shows that the PHA algorithm only partially removes the false warming signal from increasing urbanization.
- Homogenization caused the average warming trend of all stations to almost double (+89%), from +0.036 to +0.067 deg. C every decade.I can’t explain this. This may be due to real effects of instrumentation changes, time-of-observation adjustments (TOBS), unintended artifacts of the PHA algorithm, or a combination of the three.
Does This Mean Recent Warming in the US Is Ignoring It?
Maybe not. Although it suggests a warming trend since 1895, if we examine the most recent warming period (say, since 1961 … a date I chose arbitrarily), we find a much stronger warming trend.
Note that the GHCN trend since 1961 is almost identical from the raw (+0.192 C/decade) to the homogenized data (+0.193 C/decade). The average UHI warming trend is only about 13% of the raw GHCN trend, and 10% of the homogenized trend, indicating that little of the GHCN warming trend can be attributed to the increase in population density.
But there is still a signal of urbanization in the raw data and the adjusted data, as shown by the non-zero regression slopes. One possible interpretation of this result is that the homogenization algorithm is distorting the trend of the station, and if we can use the raw GHCN data as a more accurate representation of reality, then intercept the regression +0.10 deg. C/decade is the best estimate of the average warming trend of all stations if no station has population growth. That is smaller than 50% of the warming trend of the homogenized data +0.192 deg. C/decade.
What Does It All Mean?
First, there is evidence supporting the “urban mixing” hypothesis of Katata et al., with the homogenization algorithm inadvertently mixing the characteristics of urban stations into rural temperature data. This appears to increase the average temperature trend of all stations.
Second, homogenization appears to remove only about 50% of the UHI signal. Even after homogenization, the GHCN temperature trend tends to be higher for stations with large population growth, lower for stations with small population growth. There is some evidence that rural stations actually have only about 50% of the average warming of all US stations, which is consistent with Anthony Watts’ estimate based on a restrictive analysis of only the best stations.
The results show there is now an additional reason to distrust official temperature trends reported for US weather stations. They are, on average, very warm. How much? That remains to be determined. Our method provides the first method (that I know of) of estimating the effect of urban warming independently over time, even in an average sense (that is, accurate for the average over many stations, but the accuracy at individual stations is unknown) . As my career winds down, I hope others will outgrow this type of analysis.
(To see the number of UHI signals in various calendar months around the world from 2023, here are high-resolution images: Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec. More details about our method, along with a link to a monthly ArcGIS format file of the global UHI grid since 1800 (Version 0) is in my blog post from November, 2023.)