Home/BlogIn Part I I showed the Landsat satellite-based measurements of urbanization around the Global Historical Climate Network (GHCN) land temperature-monitoring stations. Virtually all of the GHCN stations have experienced growth in the coverage of human settlement “built-up” (BU) structures.
As an example of this growth, here is the 40-year change in BU values (which range from 0 to 100%) at 1 km spatial resolution over the Southeast United States.
How has this change in urbanization been expressed at the GHCN stations distributed around the world? Fig. 2 shows how urbanization has increased on average across 19,885 GHCN stations from 20N to 82.5N latitude, at various spatial averaging resolutions of the data.
NONE of the 19,885 GHCN stations experienced negative growth, which is not that surprising since that would require a removal of human settlement structures over time. In all of the analysis that follows, I will be using the 21×21 km averages of BU centered on the GHCN station locations.
So, what effect does urbanization measured in this manner have on GHCN temperatures? And, especially, on temperature trends used for monitoring global warming?
While we all know that urban areas are warmer than rural areas, especially at night and during the summer, does an increase in urbanization lead to spurious warming at the GHCN stations that experienced growth (which is the majority of them)?
And, even if it did, does the homogenization procedure NOAA uses to correct for spurious temperature effects remove (even partially) urban heat island (UHI) effects on reported temperature trends?
John Christy and I have been examining these questions by comparing the GHCN temperature dataset (both unadjusted and adjusted [homogenized] versions) to these Landsat-based measurements of human settlement structures, which I will just call “urbanization”.
Here’s what I’m finding so far.
The Strongest UHI Warming with Urbanization Growth Occurs at Nearly-Rural Stations
As Oke (1973) and others have demonstrated, the urban heat island effect is strongly nonlinear, with (for example) a 2% increase in urbanization at rural sites producing much more warming than a 2% increase at an urban site. This means that a climate monitoring dataset using mostly-rural stations is not immune from spurious warming from creeping urbanization, unless there has been absolutely zero growth.
For example, Fig. 3 shows the sensitivity of GHCN (absolute) temperatures to increasing urbanization in various classes of urbanization, based upon well over 1 million station pairs separated by less than 150 km.
By far the greatest sensitivity to a change in urbanization in Fig. 3 is in the 0-2% (nearly rural) category. We also see in Fig. 3 that the homogenization procedure used by NOAA reduces this effect by only 9% averaged across all seasons, and by even less (2.1%) in the summer season.
If we integrate the sensitivities in Fig. 3 from 0 to 100% urbanization, we get the total UHI effect on temperature (Fig. 4).
The temperature data used here is the average of the daily maximum and minimum temperatures ([Tmax+Tmin]/2), and since almost all of the urban heat island effect is in Tmin, the temperature scale in Fig. 4 would be nearly doubled for the Tmin UHI effect.
The black curve in Fig. 4 is a square-root relationship, which seems to match the data reasonable well for most of the GHCN stations (which are generally less than 30% urbanized). But this is not nearly as non-linear as the 4th root relationship Oke (1973) calculated for some eastern Canadian stations, using population data as a measure of urbanization.
But what I have shown so far is based upon spatial information (the difference between closely-spaced stations). It does not tell us whether, or by how much, spurious warming exists in the GHCN temperature trends. To examine this question, next I looked at how the NOAA homogenization procedure changed station trends as a function of how fast the station environment has become more urbanized.
NOAA’s homogenization produces a change in most of the station temperature trends. If I compute the average homogenization-induced change in trends in various categories of station growth in urbanization, we should see a negative trend adjustment associated with positive urbanization growth, right?
But just the opposite happens.
First let’s examine what happens at stations with no growth in urbanization. In Fig. 5 we see that the 881 stations with no trend in urbanization during 1975-2014 have an average 0.011 C/decade warmer trend in the adjusted (homogenized) data than in the unadjusted data. This, by itself, is entirely possible since there are time-of-observation (“Tobs”) adjustments made to the data, adjustments for station moves, instrumentation types, etc.
So, let’s assume that value at zero growth in Fig. 5 represents what we should expect for the NON-urbanization related adjustments to GHCN trends. As we move to the right from zero urbanization growth in Fig. 5, stations with increasing growth in urbanization should have downward adjustments in their temperature trends, but instead we see, for all classes of growth in urbanization, UPWARD adjustments instead!
Thus, it appears that NOAA’s homogenization procedure is spuriously warming station temperature trends (on average) when it should be cooling them. I don’t know how to conclude any different.
Why are the NOAA adjusments going in the wrong direction? I don’t know.
To say the least, I find these results… curious.
OK, so how big is this spurious warming effect on land temperature trends in the GHCN dataset?
Before you jump to the conclusion that GHCN temperature trends have too much spurious warming to be relied upon for monitoring global warming, what I have shown does not tell us by just how much the land-average temperature trends are biased upward. I will address that in Part III.
My very preliminary calculations so far (using the UHI curves in Fig. 4 applied to the 21×21 km urbanization growth curve in Fig. 2) suggest the UHI warming averaged over all stations is about 10-20% of the GHCN trends. Small, but not insignificant. But that could change as I dig deeper into the issue.
