I’ve been spending recent months applying our novel methodology of quantifying the urban heat island (UHI) effect on surface air temperature, now using Landsat-based Impervious Surface (IS) cover fraction as a proxy for urbanization. This is an adaptation of our published research using population density (PD) as a proxy for urbanization, in which we showed that about 60% of the U.S. warming trend since the late 1800s in urban and suburban areas could be attributed to increases in population density. We used non-homogenized (raw) GHCN temperature data in that study; it remains unknown to what extent homogenization procedures implemented by NOAA, Berkeley BEST, et al. have removed this spurious warming effect.

One important aspect of the population density-based research was that the UHI effect on U.S. warming trends largely disappeared after about 1960. We used population density for that study because there are global gridpoint datasets of PD at approximately 10 km spatial resolution going back centuries. So, it was a data availability choice.

But the more physically direct proxy for urbanization in the context of the UHI effect is how much of the surface is covered by impervious surfaces (mainly roads, parking lots, buildings, etc). There are now Landsat-based datasets of IS coverage over the U.S. at high spatial resolution (~30 m) but only since 1985 when Landsat data quality was sufficient for such retrievals. This post addresses some results using those IS data. Here’s an example of IS data for the NYC area in 2024:

Specifically, I’m going through the top major Metropolitan Statistical Areas (MSAs) ranked by total population to quantify the average summertime (JJA) UHI impact on daily maximum temperature (Tmax) and minimum temperature (Tmin). I’m computing these effects separately for excessively hot days (~97th percentile) versus non-excessively hot days, which is yielding some interesting results. The analyses are based upon all available GHCN daily data during the summers of 1985 through 2025 within 40 to 100 km of the approximate centroidal location of the major metropolitan areas.

The Surprising (to me) Impact of Elevation, Nighttime Watering, and Daytime Ocean- and Lake-Breezes

Elevation

One thing I enjoy about analyzing large datasets is when I find something that surprises me… even when it shouldn’t have surprised me. The first effect was elevation. We all know that temperature decreases with height in the troposphere. This is why other UHI studies have required urban thermometer locations to be at elevations not very different from the rural locations. The “gold standard” requirement has been no more than 10 m or 30 m difference in elevation.

The problem with this standard is that it greatly restricts the number of available GHCN stations being analyzed. Since the UHI effect is often not much more different from station-specific biases due to other factors, one needs as many stations as possible to beat down the noise and extract the UHI signal. I have been using a rather loose 100 m to 250 m, but I gradually realized this was causing a bias in the results.

Why a bias, rather than just elevation difference-related noise? As I went down the list of the top U.S. metropolitan areas I realized that virtually all of them have something in common: they are at average elevations lower than the surrounding rural areas. This makes sense historically since major cities were originally developed next to major water bodies to factilitate transportation: the ocean, major rivers, and large lakes, which are all at lower elevations than their surroundings. This means that a portion of what we perceive to be the urban heat island effect is often due to differences in elevation. Sometimes there isn’t a major water body (e.g. Las Vegas), but for several practical reasons cities are seldom built in the mountains; they are instead in the low-lands.

So, I implemented a multiple regression procedure to separate out the impact of elevation from impervious surface cover in my calculations. This allows me to use all available stations, no matter their elevation, which helps to beat down the noise from other, non-UHI effects on measured air temperatures.

Nighttime Watering

I also found that most of the western U.S. cities have curious UHI effects, expecially during excessively hot days. Most of the U.S. West is characterized by summertime drought as a persistent feature of the weather there. I am now pretty sure that in many of these cases the curious results are due to nighttime watering of vegetation, which increases during excessively hot days.

Ocean and Lake Breezes

Several major cities experience significant daytime ocean breezes (e.g. Los Angeles) or lake breezes (e.g. Chicago). This acts against the urban heat island warming. As we will see, in the case of Los Angeles the cooling sea breeze almost totally dominates over any UHI warming.

Some Major Metropolitan Area Results

My methodology uses all available GHCN station pairs available on each summer day for the years 1985 through 2025. For each station pair, I compute the temperature differences (Tmax and Tmin, separately), as well as the differences in 1×1 km average impervious surface coverage centered on those station locations (I also looked at 2×2, 5×5, and 10×10 km results). This is done for all station pairs within 40 km to 100 km (city-dependent) of the approximate centroid of the MSA being considered (in the case of NYC, I chose Central Park). I then group all of these station-paired data into 7 classes of 2-station average IS coverage, which allows me to examine any nonlinearities in the UHI-vs-IS relationships. For each class, I regress the temperature differences against the IS differences to get an average dT/dIS (regression slope) value. These 7 slopes are then integrated across IS to arrive at curves of UHI temperature impact versus IS.

An important feature of the method is that I don’t have to categorize a station as “rural” or “urban”, as most other UHI studies have done. As seen in Fig. 1 (above) there is a continuum of urbanization as quantified by IS coverage from 0% (wilderness) to 100% (complete coverage by roads, parking lots, buildings, etc.)

New York City-Newark-Jersey City MSA

The New York-Newark-Jersey City MSA is the most populous in the U.S., with 6% of the U.S. population residing there. Fig. 2 shows the resulting average UHI effects across this MSA on Tmax and Tmin, and for excessively-hot days vs. not excessively hot days.

(It is important to point out that these results should not be interpreted as necessarily representing inner-city NYC vs. surrounding rural areas. They are the average results for all available station pairs found within 60 km of Central Park, thus are for stations generally not in downtown NYC. Instead, they provide an average picture of how urbanization affects air temperatures, on average, across the entire metropolitan region.)

The first thing we see in Fig. 2 is that the UHI warming effects are much larger on Tmin than on Tmax, which many others have found.

Secondly, we see that excessively hot days have a somewhat stronger UHI warming effect at the most urbanized locations (largest IS values). But for Tmax on non-excessively hot days there is evidence of the “urban cool island” effect, which others have studied and published results on. This is a natural consequence of impervious surfaces conducting heat down into the sub-surface compared to natural land (and vegetation) surfaces, which causes a time lag in the diurnal temperature response.

Los Angeles-Long Beach-Anaheim MSA

We need only go to the 2nd most populous MSA (Los Angeles) to see that the temperature changes in urban areas are not always due to warming from urbanization. This is shown in Fig. 3.

In this case we see a huge daytime cooling effect on Tmax in urban areas, which I assume is due to the persistent daytime sea breeze in the LA basin during summer. The effect is also seen to a lesser extent in Tmin for excessively hot days. I don’t know whether this is due to stronger and more persistent sea breezes on excessively hot days, or due to more nighttime watering of vegetation during those days, or some combination of both.

At this point you might be wondering, how can the hottest days have cooler urban temperatures? This is where I have to explain how I classify “excessively hot days”. Because there are so many GHCN stations within 40 km of downtown LA, there are days when some stations exceed their 97th percentile hottest temperature and other stations do not. So how do we decide which days are “excessively hot” for the metropolitan region as a whole? I calculate for each date in the summers of 1985-2025 how many stations exceed their 97th percentile threshold. I then compute the average daily temperature across those stations. For LA, it turns out at least 12 stations exceeding their 97th percentile temperature threshold are required in order for approximately 3% of the dates to be categorized as “excessively hot”, thus providing a 97th percentile threshold for the whole MSA region. I then use that 12-station minimum, applied to Tmax (not Tmin), to decide which dates are “excessively hot”.

I am finding that most of the major cities in the western U.S. have reduced UHI heating (and like LA, even cooling) during daytime and nighttime on excessively hot days. In many cases I believe this is due to watering of vegetation, which for every city I have checked, Grok says that city has more water usage during the nighttime hours on excessively hot days. For example, here are the results for Portland-Vancouver-Hillsboro, the 24th most populous MSA in the U.S; note how the fairly strong UHI warming effect on Tmax and Tmin is reduced on the hottest days, especially at night when most watering occurs:

For the bottom curve in Fig. 4 (nighttime Portland temperatures on excessively hot dates), one might even imagine the maximum cooling effect from more watering is in the suburbs (IS less than 20-30%), but then switching to warming in the most urban areas (IS over 50%), presumably due to differences in areal coverage by vegetation being watered.

I am through about two dozen of the 50 most populous metropolitan areas I want to include results for as part of a paper we are preparing for submission to the journal Urban Climate. Since those 50 MSAs include over 50% of the U.S. population, chances are good your city or town will also be included.