Review of Spencer's New Paper on Urban Heat Islands
A new paper[1] by Spencer and Christy was published on urban heat islands (UHI), and I'd like to clarify what it says and what can actually be claimed as a result of it. The reason why has to do with a recent blog post from WUWT that claims the paper shows that 65% of global warming is due to UHI warming effects, rather than increasing greenhouse gas concentrations. Chris Rotter at WUWT says,
A new study from the University of Alabama in Huntsville addresses the question of how much the Urban Heat Island (UHI) effect is responsible for the higher temperatures at weather stations across the world. Dr. Roy Spencer and Dr. John Christy have spent several years developing a novel method that quantifies, for the first time, the average UHI warming effects related to population density. Their finding: no less than 65% of “runaway global warming” is not caused by our emissions of carbon dioxide, but by the urbanization of the world.
There's very little in this that resembles what the paper actually claims. The first clue to Rotter's dishonesty could come from just reading the title of Spencer's paper: "Urban Heat Island Effects in U.S. Summer Surface Temperature Data, 1895–2023. So this paper evaluates the impact of UHI on CONUS summer temperatures, not annual global temperatures. But if you read the abstract of the paper, it gets worse. The abstract is below.
A novel method is described for quantifying average urban heat island (UHI) warming since 1895 in contiguous U.S. (CONUS) summer air temperature data. The method quantifies the sensitivity of Global Historical Climatology Network (GHCN) station raw temperature to station-centered population density (PD). Specifically, closely spaced station pair differences in monthly raw (non-homogenized) Tavg (the average of daily maximum and minimum temperature) and PD are sorted by station pair average PD into six PD classes, and linear regression estimates of the temperature sensitivity to population density change (dTavg/dPD) are made for each class for historical periods ranging from 1 to 21 years in length. Every one of the resulting six sensitivity relationships in each of 22 historical periods from 1880 to 2020 are found to be positive, and their magnitudes allow construction of station-average urban heat island temperature (Tuhi) curves as a function of population density. When applied to the history of population changes at each CONUS station location (1895–2023) and grouped into four categories of station population density, the resulting Tuhi warming trends range from 8% of observed TAVG warming for the most rural category of stations to about 65% of observed warming for suburban and urban categories. Across all stations the UHI warming amounts to 22% of the observed raw GHCN warming trend, (+0.016 versus +0.072 °C decade−1). The method provides an independent way to quantify station-average UHI warming over time.
The details in the abstract alone show that Spencer's paper bears little resemblance to the blogpost, and the actual results simply do not have a significant impact on our understanding of global warming. Let's clarify what the paper is about:
- The paper only examines US temperatures, so only 2% of the globe.
- The paper only evaluates summer US temperatures, so only 25% of the year.
- The paper found that UHI explains only 22% of CONUS summer warming, not 65%.
- The paper reported biases to groups of station data, but these are not weighted to the fraction of surface area represented by each group.
- The paper used only raw temperatures before homogenization is applied. Since homogenization effectively removes urbanization biases, this paper cannot constitute an argument that bias currently affects homogenized datasets of CONUS temperatures.
So the WUWT summary bears little resemblance to the actual paper. Yet I suspect the paper will still be misused to suggest that CONUS temperature trends are being significantly affected by urbanization biases, and I think it's pretty clear that this paper does not support that claim. There are at least three reasons why I think this is the case: 1) they used raw, not homogenized data, 2) their 22% figure was not area-weighted, and 3) their "rural" subset of stations includes many stations that stopped being rural by 2020.
Use of Raw instead of Homogenized Data
Spencer admits in his paper that his analysis doesn't determine how much urbanization biases affect NOAA's homogenized datasets like nClimDiv. They write, "It is beyond the scope of this paper to determine how much (if any) urban warming remains in the adjusted (homogenized) GHCN data; here we will compute UHI effects from the raw (unadjusted) version of the dataset." In other words, this paper doesn't argue that any warming in homogenized datasets for CONUS temperatures is due to the urban heat island effect. Even if we accept everything the paper claims, it only quantifies the bias in raw temperature data that is subsequently removed during bias correction/homogenization.Spencer's blog post about this paper contains a further admission along these lines. He points out that while the paper was undergoing peer review, they were told to use raw data instead of homogenized data because homogenization removes the bias that Spencer was looking to find. Here's an excerpt:
What Does This Mean for Urbanization Effects in the Official U.S. Temperature Record?
That’s a good question, and I don’t have a good answer.
One of the reviewers, who seemed to know a lot about the homogenization technique used by NOAA, said the homogenized data could not be used for our study because the UHI-trends are mostly removed from those data. (Homogenization looks at year-to-year [time domain] temperature changes at neighboring stations, not the spatial temperature differences [space domain] like we do). So, we were forced to use the raw (not homogenized) U.S. summertime GHCN daily average ([Tmax+Tmin]/2) data for the study. One of the surprising things that reviewer claimed was that homogenization warms the past at currently urbanized stations to make their less-urbanized early history just as warm as today.
I doubt very seriously that they were "forced" by the reviewer to use raw data and not use homogenized data. There's no reason why Spencer couldn't have done a comparative analysis of both raw and homogenized data to discover the extent to which homogenization removed the bias discovered in the raw data. Yet Spencer seems to believe that a bias remains in the homogenized data:
That reviewer of the paper said most of the spurious UHI warming effect has been removed by the homogenization process, which constitutes the official temperature record as reported by NOAA. I am not convinced of this, and at least one recent paper claims that homogenization does not actually correct the urban trends to look like rural trends, but instead it does “urban blending” of the data.So if that's the case, why not test it? Of course, in many ways, this has actually already been done. Wickham et al 2013[2] isolated the most rural stations in the Berkeley Earth network and compared this subset to all stations and found that these most rural stations are warming at least as rapidly as all stations; this effectively rules out the possibility that spurious UHI warming effects are contributing to global land warming trends.
 |
| Wickham's Analysis of All vs Very Rural Stations |
If homogenization caused an "urban blending" of the data, why are the most rural stations warming more rapidly than all stations? Urban and rural areas warm at about the same rate; the bias is added where urbanization is taking place. So even if there is an "urban blending" of temperature trends in homogenized data in those areas undergoing urbanization, given that urban and rural trends are have essentially the same trends, how is this a criticism? There is simply no evidence that UHI is adding any spurious warming trends to homogenized datasets, a point that can is reinforced with CONUS temperatures below.
No Area-Weighting of Station Data
This paper is only reporting bias affecting stations from rural, peri-rural, suburban and urban areas, but these are not area-weighted by the amount land area in each category. The graph below shows the quantifications of UHI biases affecting raw summer temperatures for various categories of stations.
If we accept these numbers, we can calculate the effect of UHI on raw temperature data if we know the fractions of land surface area covered by the the four categories. The following should be noncontroversial from the data I've been able to find:
- Rural (Tr) - 87.4% of CONUS
- Peri-rural (Tp) - 9.6% of CONUS
- Suburban (Ts) - 2% of CONUS
- Urban (Tu) - 1% of CONUS
Some may quibble with these numbers a little, but I think we can all agree these are in the ballpark. So for any UHI-free warming rate (Tf), we should be able to calculate the warming in raw data with UHI effects (Tuhi) with the following equation:
Tuhi = [0.874*Tf*1.082] + [0.096*Tf*1.417] + [0.02*Tf*1.637] + [0.01*Tf*1.669]
For any value of Tf, Tuhi is 9.8% more rapid, so the area-weighted effect of UHI on summer CONUS raw temperature data is about 10%. Since this is only summer (JJA) data, when the UHI effect is the strongest, this value would be even smaller using annual means instead of summer means. And this is again before homogenization procedures remove urbanization biases from datasets like nClimDiv.
In another post I shared a "sniff test" for how much urban warming would need to take place in order to explain various levels of presumed bias affecting GMST data. I'd like to do the same here using nClimDiv (homogenized) data for summer (JJA) temperatures. The nClimDiv dataset uses the same GHCN-daily data used by this study, so it should be a good way to evaluate to what extent nClimDiv's gridding and homogenization procedures remove the biases quantified by this paper. Using this data, summer mean temperatures (Tus) since 1970 have increased at a rate of 0.24°C/decade. Let's lump together all the non rural areas (urban, suburban and peri-rural), making up 12.6% of CONUS, and see how rapidly these non-rural (Tnr) areas must warm to explain a Tus = 0.24°C/decade. For this, Tus = (12.6Tnr + 87.4Tr)/100 = 0.24°C/decade (1.3°C global warming since 1970), and we can solve for Tnr as
Tnr = (100Tus - 87.4Tr)/12.6
I'll use the 41.7% figure for peri-rural stations in the above chart to favor proponents of the UHI effect; this will low-ball the Tnr warming required to achieve Tus. Using this data, rural areas with no urbanization effects would warm at a rate of Tr = 0.24 * 0.583 = 0.14°C/decade (0.79°C rural warming for 1970-2024). This means Tnr = (100*.24 - 87.4*0.14)/12.6 = 0.95°C/decade (5.23°C urban warming for 1970-2024). Spencer's results would indicate that non-rural warming has progressed at a rate at least 6.7x more rapidly than rural areas unaffected UHI. This means we would expect the map below of ERA5 data to show deep red areas showing ~5°C warming in non-rural areas and metropolitan areas, and the rest of the country showing ~0.8°C warming. We should see major hot spots surrounding cities in the US. Instead, you can't pick out a single metropolitan area in the map below. It would appear that gridding and homogenization procedures are removing the urbanization biases from CONUS summer temperatures pretty effectively. No hint of the UHI effect can be seen in ERA5 data.
Poor Definition of "Rural" Stations
I think the paper's definition of "rural" station is misleading. Spencer found that UHI added 8% to the trend to rural stations. I question whether this is actually the case, since the definition of the UHI effect is that urban areas are warmer than rural areas, meaning pretty much by definition, Tuhi should be seen as Tu - Tr. So the warming at the "most rural" stations in 2020 should be considered free of UHI effects. If they are affected by UHI, then they aren't rural. In fact, I think this can be confirmed by this graph from the paper.
As I read this, the cumulative population density growth of stations labeled "rural" in the above graph stopped being rural around 1940, and and the cumulative growth of "peri-rural" stations stopped being peri-rural around 1970. I take issue with this. These stations should have been categorized as rural or peri-rural based on their population densities in 2020, virtually ensuring that stations counted as rural have always been rural (that's how this was handled in Wickham et al 2013). I suspect this oversight accounts for the 8% UHI attribution in rural stations. I don't think this paper shows that currently rural stations are have an urbanization bias. I think it shows that urbanization has caused some stations to become peri-rural, and this again is just the kind of change that would be picked up and addressed by homogenization.
Reference:
https://www.scitechnol.com/2327-4581/2327-4581-1-104.pdf
Comments
Post a Comment