USHCN vs. Hourly Weather Stations «Roy Spencer, PhD

SUMMARY: Urban Heat Island (UHI) impacted US temperature trends in the official USHCN dataset from the NOAA 1,218 station. I contend that because of the importance of quality temperature trend calculations to national energy policy, a new dataset is required that does not depend on the USHCN Tmax / Tmin observations. I find that regression analysis applied to the hourly ISD weather data (mainly from airports) between the temperature trends of many stations and the local population density (as a UHI proxy) can be used to determine the average annoying warming trend component due to UHI remove. Using the hourly station data provides a largely USHCN-independent measure of the warming trend in the United States without the need for uncertain adjustments to the observation time. The resulting average US trend with 311 stations (1973-2020) after removing the UHI-related Spurios trend component is approximately +0.13 degrees. C / decade, which is only 50% of the USHCN trend of +0.26 C / decade. Regarding the quality of the station data, the variability between the raw data from the USHCN stations is 60% greater than among the trends calculated from the hourly data, suggesting that the USHCN raw data is of poorer quality. It is recommended that a de-urbanization of trends be applied to the hourly data (mainly from airports) in order to have a more accurate record of temperature trends in land areas like the US that have enough temperature data to generate the UHI . vs trend correction.

The Urban Heat Island: Average vs. Trending Effects

Over the past 50 years (1970-2020), the U.S. population has increased by a whopping 58%. More people mean more infrastructure, more energy consumption (and waste heat generation), and even if the population does not increase, our rising standard of living leads to a multitude of increases in production and consumption with more businesses, parking spaces and air conditioning, etc.

As T.R. Oke showed in 1973 (and many others since then) that the UHI has a significant impact on surface temperatures in populated regions up to several degrees Celsius. The additional heat is created both by waste heat and by replacing cooler, overgrown surfaces with impermeable and easy-to-heat hard surfaces. The effects can occur on many spatial scales: a heat pump placed too close to the thermometer (microclimate effect) or a large city with outward expanding suburbs (mesoscale effect).

Over the past 20 years (2000 to 2020) the population growth has been mainly in urban areas, with no average increase in rural areas. Fig. 1 shows this for 311 hourly weather station locations that have had relatively complete weather data since 1973.

This could suggest that only rural data are used to monitor temperature trends. The disadvantage is that there are relatively few station locations with a population density of less than, say, 20 people per square kilometer, so coverage of the United States would be quite poor.

What would be nice would be if the UHI effect could be eliminated on a regional basis based on how average warming trends increase with population density. (Again, this is not eliminating the average temperature difference between rural and urban areas, but rather eliminating spurious temperature trends due to UHI effects).

But is there such a relationship at all?

UHI Effects on USHCN Temperature Trends (1973-2020)

The most commonly cited data set on surface temperature used to monitor global warming trends in the United States is the US Historical Climatology Network (USHCN). The dataset contains a fixed set of 1,218 stations whose records go back over 100 years. Since most of the station data consists of daily maximum and minimum temperatures (Tmax and Tmin) measured daily at a single time, and this observation time (TOBs) changed from late afternoon to early morning around 1960 (discussion here), A TOBs-related temperature distortion occurred, the size of which is somewhat uncertain, but which still needs to be adjusted.

NOAA provides both the unadjusted and the adjusted raw data (TOBs & spatial “homogenization”). The following diagram (Fig. 2) shows how the two station temperature trends of the data sets correlate with population density. This should not be the case if UHI effects have been removed from the trends.

Fig 2. The temperature trends of the USHCN stations correlate with the population density. This shouldn’t be the case with the Urban Heat Island effect on trends removed.

Any UHI effect on temperature trends It would be difficult to remove them through the NOAA homogenization process alone. That’s because, If all the stations in a small area, both urban and rural, incorrectly warm up due to UHI effects, this signal will not be removed as it is expected for global warming as well. Adjustments to “homogenization” can theoretically lead to trends in rural and urban areas being the same. However, this does not mean that the UHI effect has been eliminated.

Instead, one must examine the data in a manner similar to Figure 2, which shows that even the fitted USHCN data (red dots), when extrapolated, still overestimate the US station mean trends (1973-2020) by about 30% a regression relationship (red dashed line, 2nd order polynomial fit) to zero population density. However, such an analysis requires many stations (i.e. large areas) to measure the average effect. It is not clear how many stations it would take to get a robust signal. The more stations are required, the larger the regional area.

US hourly temperature data as an alternative to USHCN

There are many weather stations in the US that (mostly) are not included in the USHCN set of 1,218 stations. These are the operational hourly weather stations operated by NWS, FAA and other agencies that provide most of the data that the National Weather Service reports to you. The data is contained in the ISD archive (Integrated Surface Database) for several agencies.

The data archive is quite large as it contains (up to) hourly resolution data (higher with “special” observations in changing weather) and many weather variables (temperature, dew point, wind, air pressure, precipitation) for many thousands of stations around the world. Many of the stations (at least in the US) are located at airports.

In the US, most of these measurements and their reporting are now automated with the AWOS and ASOS systems.

This map shows all of the stations in the archive, although many of them will not have complete records for the decades of interest.

Fig. 3. Locations of ISD surface weather data that are quality controlled and stored at NOAA.

The advantage of this data, at least in the US, is that the devices are regularly serviced. When I was working in an office for the National Weather Service in Michigan that summer, there was a full-time Met Tech servicing and adjusting all weather gauges.

Since observations are (nominally) every hour on the hour, no uncertain adjustment of TOBs is required, as is the case with the USHCN’s daily Tmax / Tmin data.

The area with average population density differs significantly between the ISD stations (“hourly”) and the USHCN stations, as shown in Fig. 4.

Fig. 4. The dependence of the population density of US weather stations on the average area differs significantly between 1,218 USHCN and 311 high-quality ISD stations (“hourly”), mainly due to the measurement of the hourly data at “uninhabited” airports that are supported should flight safety.

In Fig. 4 we see that the population density in the immediate vicinity of the ITS stations in the immediate area of ​​1 km² is on average only 100 people, since nobody “lives” at the airport, but increases considerably with the average area, since there are population centers too to serve.

In contrast to this, the USHCN stations have their highest population density in the immediate vicinity of the weather station (over 400 people in the first km²), which then drops with the distance from the station location.

How such differences affect the extent of UHI-related disruptive warming trends is currently unknown.

UHI effects on the hourly temperature data

I analyzed the US ISD data for the lower 48 states for the period 1973-2020. (Why 1973? Since much of the early records are on paper and in hourly resolution, this means a lot of manual digitizing. Apparently 1973 is as far back as much of this station data was digitized and archived.)

First, I only average the temperatures of 00 UTC and 12 UTC (roughly the times of the maximum and minimum temperatures in the USA). I required that these twice daily readings be reported for at least 20 days so that a month could be considered for inclusion, and then at least 10 out of 12 months from a station to have good data for a year of data from that station stored.

For the temperature trend analysis, I then requested that 90% of the years 1973-2020 must contain data, including the first 2 years (1973, 1974) and the last 2 years (2019-2020), as end years can have a large impact on trend calculations.

The resulting 311 stations have a commonality of 8.7% with the 1,218 USHCN stations. This means that only 8.7% of the (mostly airport) stations are also included in the USHCN database with 1,218 stations, so that the two data sets are largely (but not completely) independent.

I then plotted the equivalent of Fig. 2 for the ISD stations (Fig. 5).

Fig. 5. As in Fig. 2, but for the ISD station trends (mostly airport) for the average of the daily temperatures of 00 and 12 UTC. Where the regression lines intersect the zero population axis is an estimate of the U.S. temperature trend over the 1973-2020 period, with spurious UHI trend effects removed.

For the linear fit to the data, we can see that extrapolating the line to zero population density gives an average warming trend of 311 stations of +0.13 degrees. C / decade.

Significantly, this is only 50% of the official TOBs-adjusted, homogenized average trend of +0.26 C / decade of the USHCN 1,218 stations.

It is also significant that this 50% reduction in the official US temperature trend is very close to what Anthony Watts and coworkers got in their 2015 analysis using the best-located USHCN stations.

I also include the polynomial fit in 5 because my use of the fourth root of the population density is not meant to perfectly capture the nonlinearity of the UHI effect and it is expected that some nonlinearity will be preserved. In this case, the extrapolated warming trend is close to zero for a population density of zero. For the current discussion, however, I will use the linear fit in Fig. 5 conservatively. (The logarithm of population density is also sometimes used, but it doesn’t behave well when the population approaches zero.)

Evidence that the raw ISD station trends are of higher quality than UHCN’s is in the standard deviation of these trends:

Std. Dev. Of 1,218 USHCN (raw) trends = +0.205 Degree C / decade

Std. Dev. Of 311 ISD trends (“hourly”) = +0.128 Degree C / decade

Thus, the variation in the USHCN raw trends is 60% greater than the variation in the hourly station trends, suggesting this The airport trends have less time-changing disruptive temperature influences than the USHCN station trends.

Conclusions

For the period 1973-2020:

1. The homogenized USHCN data still has false warming influences related to UHI (Urban Heat Island) effects. This has exaggerated the global warming trend for the US as a whole. The size of this interfering component is uncertain due to the black box nature of the “homogenization process” applied to the raw data.
2. An alternative analysis of US temperature trends using a largely independent data set from airports suggests that the US UHI adjusted mean warming trend (+0.13 ° C / decade) may only be 50% of the official USHCN station trend trend (+0.26 ° C) is C / decade).
3. USHCN raw trends show 60% more variability than raw airport trends, indicating a higher quality of routinely maintained airport weather data.

Future work

This is an extension of the work I started about 8 years ago but never finished. John Christy and I are discussing using results based on this methodology to create a new US surface temperature dataset that is updated monthly.

I’ve just outlined the basics above. You can do similar calculations in subregions (I find the western US results are similar to the eastern US results). In addition, the results would likely be seasonal. In this case it should be calculated by calendar months.

Of course, the methodology could also be applied to other countries.