The urban heat island effect has not yet been eliminated from the official GHCN warming trend – what are your thoughts on this?

From Dr. Roy Spencer's Global Warming Blog

Author: Dr. Roy W. Spencer

Our paper (co-authored by John Christy and Danny Braswell) on calculating the urban heat island effect as a function of population density (PD) is now in the final stages of review after a third round of edits, and I'm hoping it will be published soon Accepted for publication. So far, I have only used Tavg data (average of daily maximum and minimum temperatures) when developing and testing this method, and this article has only used continuous US summer data (June, July, August), which That's what I'm going to discuss here.

This approach allows us to calculate UHI trends using a global gridded PD dataset dating back to the 1800s. These UHI trends can then be compared with the GHCN station temperature trends. If I do this for all US GHCN stations that have at least 120 years of complete monthly (June, July, or August) data out of 129 potential years between 1895-2023, the graph below shows some interesting results. (I started with “raw” data so we could examine how homogenization changes the results.)

1. The greater the population growth at a station, the greater the observed warming trend. This is very convincing evidence that the raw GHCN data have significant UHI effects affecting calculated trends (probably not surprising here). Note that the UHI temperature trend is on average 66% of the original temperature trend.
2. Fitting a regression line clipping zero data shows that On average, sites with no population growth had no warming trend. While this may lead some to conclude that there was no net warming in the United States between 1895 and 2023, it is important to remember that these are raw data and have not been adjusted for time of observation (TOBS) changes or instrument type changes. These changes may result in a trend toward lower temperatures for most or all sites.

Since most rural sites (many of which have little population growth) are in the western United States, and there may be differences in actual temperature trends between the eastern and western US, let's see if we only examine the eastern US (Ohio to the Florida Peninsula, east):

This suggests that the eastern US has similar characteristics to the US as a whole, with a regression line intercept of zero (again) indicating that those sites with no population growth have (on average) no warming trend in the raw GHCN data. But now, surprisingly, the average UHI trend exceeds 95% (!) of the original station trend which seems to indicate that essentially all reported warming in the eastern US from 1895 to 2023 is due to urbanization effects… If there were no systematic bias in the raw Tavg data, this would cause these trends to be low. Also, as will be discussed below, this is the period 1895 to 2023…the results are somewhat different in recent decades.

Homogenization of GHCN data has had some strange effects

Next, let's see how the adjusted (homogenized) GHCN temperature trend compares to the UHI warming trend. Recall that NOAA's Pairwise Homogenization Algorithm (PHA) used to create the “adjusted” GHCN data set (which is the basis for official temperature statistics from the government) identifies and adjusts individual sites by comparing their temperatures. Time step change time series of a site versus time series of surrounding sites. If we plot the adjusted data trend alongside the original data trend, the graph below shows some strange changes.

Here is the effect of homogenization on the raw temperature data:

1. Stations with no population growth (no warming trend on average) now have a warming trend. I can't explain this. It may be an “urban hybrid” artifact of the PHA algorithm discussed by Katata et al. (2023, and references therein), homogenization does not adjust urban stations to “look like” rural stations, but rather tends to eliminate differences between adjacent stations, causing urban influences to “bleed through” to rural stations .
2. Sites with larger population growth have less warming trends. This is the intended effect of homogeneity.
3. The UHI effect still exists in the homogenization trend, but it has been reduced by about 50% compared with the original trend. This shows that the PHA algorithm only partially eliminates the false warming signals caused by accelerated urbanization.
4. Homogenization results in an almost doubling of the station-wide average warming trend (+89%) from +0.036 to +0.067 degrees. C every ten years.I can't explain this. This may be a practical effect due to instrument changes, time of observation (TOBS) adjustments, unexpected artifacts of the PHA algorithm, or some combination of the three.

Does this mean that near-term warming in the United States is negligible?

Maybe not. While it does suggest something is wrong with the warming trend since 1895, if we examine the more recent warming period (e.g., since 1961…a date I chose arbitrarily) we find a significantly stronger warming trend .

Note that since 1961, the GHCN trends for the original data (+0.192 C/decade) and the homogeneous data (+0.193 C/decade) are almost identical. The average UHI warming trend is only about 13% of the original GHCN trend and 10% of the homogenized trend, indicating that the GHCN warming trend can hardly be attributed to the increase in population density.

However, as shown by the non-zero regression slope, there is still an urbanization signal in the original and adjusted data. One possible explanation for these results is that the regression intercept of +0.10 degrees is if the homogenization algorithm distorts the site trends, and if we can use the raw GHCN data as a more accurate representation of reality. If no station has any population growth, then C/decade becomes the best estimate of the average warming trend across all stations. This is just above 50% of the homogeneous data warming trend of +0.192 degrees. C/decade.

What does this all mean?

First, there is evidence to support Katata et al.'s “urban mixing” hypothesis, whereby the homogenization algorithm inadvertently mixes urban site characteristics into rural temperature data. This appears to increase the site-wide average temperature trend.

Second, homogenization only seems to remove about 50% of the UHI signal. Even after homogenization, sites with greater population growth tended to have higher GHCN temperature trends, while sites with less population growth tended to have lower GHCN temperature trends. There is evidence that truly rural sites are only warming about 50% of the average warming across all sites in the United States, which is consistent with estimates by Anthony Watts based on an analysis of only those best-located sites.

These results suggest there is now more reason to distrust official temperature trends reported by U.S. weather stations. On average, they are too hot. How many? This has yet to be determined. Our approach provides the first (to my knowledge) independent estimate of urban warming effects over time, albeit in an average sense (i.e., it is accurate averaged over many sites, but not across individual The accuracy of the site is unknown). As my career winds down, I hope others will expand on this type of analysis in the future.

[To see what the total UHI signal is in various calendar months around the world as of 2023, here are the hi-res images: Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec. More details of our method, along with links to monthly ArcGIS-format files of global UHI grids since 1800 (Version 0) are contained in my blog post from November, 2023.]

Relevant