Declines in hot and cold daily temperature extremes in the conterminous US, 1899–2025

Of the high-impact weather events that disrupt human thriveability, extremes of temperature are significant. Indeed, the ability of any surface-dwelling species to live productively is often constrained by these near-surface air temperature extremes which in essence create the bounding limits for survival. With the increase in infrared-absorbing greenhouse gases (GHGs) there is a concern that such temperature-related extremes may be drifting to a regime in which more frequent and intense hot events are being experienced (e.g., Seneviratne et al. 2021).

To test claims that changes in the frequency and intensity of extremes in temperature have occurred, a relatively long-term observational dataset is necessary and one in which (a) the observations are consistently observed throughout the period of interest, (b) external (i.e., non-climatic) changes around the stations are minimal, and (c) a large number of stations are available to reduce the noise due to the inevitable errors that inhabit the observational record.

This study reports on a dataset of daily maximum temperatures in the warm season (May-Sep) and minimum temperatures in the cold season (Dec-Mar) for specific stations within the conterminous US (CONUS). All values utilized are actual, observed temperatures (bias corrected when needed for merging) without any spatial or temporal interpolation or other types of homogenization methods applied to the station observations. Area-wide averages are calculated from these values by filling a 0.5° x 0.5° grid into which station metrics are interpolated using a 1/d² weighting scheme where d is the distance from the center of the gridbox to the individual station. An average year will see 97.6% of the CONUS represented.

The study period begins with the US winter of Dec 1898-Mar 1899 and ends with the summer of 2025. This first winter was selected because this season included the mid-February 1899 Great Arctic Outbreak that set more daily record low temperatures in the CONUS than any other event since that time (Kocin et al. 1988). This was also near the beginning of the effort to standardize US weather observations as stations were being established by the federal (civilian) US Weather Bureau which was instituted in 1890 and began the general operations of data collection, printing and archiving within a few years.

With such a dataset many questions about extreme events may be answered. Such questions as, when did the hottest or coldest temperatures occur? What was the magnitude of the hottest or coldest temperature relative to the expected value in each year? What is the occurrence of continuous runs of hot or cold days over a specific time-frame (i.e., heat or cold waves)? Our goal is to construct time series to answer questions about whether changes over time may be detected. We shall discuss and draw conclusions from the answers to these questions regarding long-term changes which will include some information on systematic, non-climatic influences which likely have influenced the results to a minor extent. We shall close the discussion section by examining claims about extreme temperatures made by the National Climate Assessment 5.

The United States Historical Climate Network (USHCN) was established in the 1980s by today’s National Center for Environmental Information (NCEI) to provide a “climate” dataset to examine potential changes over decades (Quinlan et al. 1987; Karl et al. 1990). The project selected those stations with (a) the minimum number of station-moves, (b) a complete or nearly complete record (i.e., few gaps), and (c) utilization of standard equipment and procedures. In many cases, a particular station’s record was not complete, but could be made so by “threading” a nearby station’s data into the record – a station with highly correlated temperatures indicating that it was experiencing the same weather regimes as the original station. A total of 1,218 stations were selected by NCEI as meeting the requirements noted above. Green River Aviation, Utah and Binghamton, New York were official US Weather Bureau (USWB) stations whose observations were excellent but unfortunately not digitally available and require further keying-in from the original “1001” forms. Figure 1 displays the stations and the nine regions to be examined later.

The alternative text for this image may have been generated using AI.

Full size image

Location of stations with “complete” time series (black dots, 1,211) and incomplete (red circles, 7)

The observed metrics of interest here are daily maximum temperatures (T_Max) for May-Sep and daily minima (T_Min) for Dec-Mar which represent the hottest and coldest temperatures. Because the fundamental metric is measured in integer degrees Fahrenheit (°F) we shall depart from conventional scientific notation and use °F as our basic unit with Celsius (C) in parentheses.

The most time-consuming portion of this project was devoted to completing the record for each of the 1,218 stations. This included identifying gaps, extending records for stations which had closed (almost half) and extending several time series back-in-time to Dec 1898.

This study will utilize the unaltered daily data from these stations and should not be confused with the often-used monthly-averaged USHCN temperature datasets which have had adjustments applied to account for various changes over time (e.g., Menne and Williams 2009; Menne et al. 2009; Williams et al. 2012).

The daily data for still-active stations are updated in the USHCN daily database and available for convenient access at https://www.ncei.noaa.gov/pub/data/ghcn/daily/hcn. To this dataset are added the observations from the NCEI-identified “threaded” stations which considerably increases the available observations for 208 of these stations and were accessed primarily from the State Climatologist Applied Climate Information System (SC ACIS v2) archive provided by the Northeast Regional Climate Center. The final dataset consists of over 40 million observations, 90.0% of which represent the NCEI-identified observations and 10.0% supplemented by this effort.

Except for the ability to access still-reporting stations through NCEI, the USHCN daily archive is mostly inactive. No new stations have been identified (threaded) for the many which have closed in the past 30 years. Thus, a major effort was undertaken here to complete those stations, selecting stations most compatible and adjusting for any bias before merging (threading) into the original station. The three basic criteria for the selection of stations were (1) high correlation, (2) small bias, (3) nearness to the USHCN station.

In many cases, all of the data for the best-match station had not been keyed-in to the NCEI Global Historical Climate Network-Daily (GHCNd) archive and so several tens of thousands of observations were manually keyed-in for this study. An example of the keying project was the North Head Weather Bureau station located at the SW corner of Washington state. This station was highly compatible with the Long Beach Experimental Station (a USHCN station which began observations later) less than 10 km away. North Head was a fully operational, US Weather Bureau (24-hr) station with excellent records. However, over 90% of the 1902–1953 data had not been keyed-in, and thus 48 years (over 13,000 entries) were keyed-in to be threaded with the Long Beach Experimental Station. Strangely, many stations in Oklahoma were missing 5, 10 or 15-year blocks of data prior to 1945. A few of these were keyed-in for completion (a single 5-year block represents 1,370 entries of May-Sep T_Max and Dec-Mar T_Min), but for others a nearby station was identified as a substitute to fill the gaps.

The time of observation issue was also addressed. Because observations for different stations were taken at different times of day, this often led to a mismatch of dates for T_Max and on rare occasions for T_Min. For example, a station taking observations at 0700 would report T_Max for the previous 24-hrs as occurring on the date when the observation was taken and not the date of occurrence (which in all likelihood was the day before). A station measuring at midnight, however, would report temperatures for the correct date. Thus, for all merges, correlations were calculated for lag + 0, lag + 1 and lag − 1 cases (e.g., Table 1). The correlation which was highest indicated whether a shift in the threaded station was needed to align with the target USHCN station. USHCN station data from NCEI were never shifted. The time of observation issue creates an undesirable artifact for certain metrics that will be discussed later.

In some cases, a few years for a few stations of the original USHCN data were replaced with nearby observations. This occurred when the USHCN station began reporting erratically before ceasing or otherwise reported obviously erroneous values. The most egregious case was found for Greenland Ranch, Death Valley California, the site for the purported 134 °F (56.7 C) on 10 July 1913 as the world record high temperature. A detailed evaluation of this measurement and others in the pre-1920 period revealed several values of excessively hot temperatures. As Spencer et al. 2025b calculated, the actual temperature on 10 July 1913 was likely 120 °F ± 2 °F (48.1 C ± 1 C) rather than 134 °F.

To be “complete,” a station was required to have a minimum of 92% of summer T_Max (winter T_Min) data available for Dec 1898 to Sep 2024. In addition, data for 1898-99 and 2021-25 were also required so that the extreme cold event of 1899 and extreme events of 2020’s would be properly represented in the dataset relative to the entire time series. The median value of data available from the 1,211 “complete” stations was 98%. The median correlation between the target station’s T_Max (T_Min) and that of the supplement was + 0.93 (+ 0.89). The median number of overlapping daily observations for the calculation of correlations and biases for T_Max (T_Min) was 540 (685).

Finally, some stations revealed some statistically diverse behavior in terms of daily temperature values. Stations along the US Pacific coast are consistently very cool in the summer as the upwelling ocean current induces cool air temperatures and low-level stratus, keeping nearby coastal stations cool. On somewhat rare occasions however, an easterly (off shore) wind develops, bringing in already warm continental air that is then heated even further by compression to near sea level as it descends from the East.

The NCEI algorithm that checks for poor data often flags legitimate data in these coastal areas as erroneous. For example, when examining the NCEI GHCNd data for the extreme high temperature in Newport Oregon, the value produced by typical searches is 94 °F (34.4 C) in 1980. However, Fig. 2 demonstrates that on 24 June 1925 Newport’s T_Max was reported as 100 °F (37.8 C). This value was clearly a spike in T_Max relative to adjacent days and so was dismissed by the algorithm, but it is indeed an accurate measurement as confirmed by surrounding stations and the direction of the wind (East). The value is listed as “-9999” in the official GHCNd archive, so it is non-recoverable to the public user, convoluting any analysis of extremes. Again, on 10 July 1961, Newport recorded another 100 °F, but the official archive flags this as “O” (“Outside limits”) so it doesn’t appear in typical queries. Several other days were above 94 °F (34.4 C) but were listed as -9999 or “O”. In examining Newport alone, it was discovered that virtually every isolated “O” and “-9999” in summer T_Max was a legitimate observation, so that more than 10 extreme high temperatures had not been accessible for traditional analysis.

The alternative text for this image may have been generated using AI.

Full size image

Daily temperatures for Newport Oregon, June 1925. Note temperature of 100 °F (37.8 C) on 24 June which is listed as “-9999” in the widely-used official GHCNd archive. Note too that the wind direction on the 24th was from the East

In a tedious task, we manually examined all of Newport’s summer missing T_Max’s and replaced them with the value from the original forms. This was also done for other Pacific Coast stations (e.g., in Washington: Seattle Urban, Aberdeen, Long Beach Exp Station, in Oregon: Tillamook, North Bend, and in California: Ft Bragg, Newport Beach Harbor, and Santa Barbara) to provide results with a more accurate accounting of the extremes. However, we cannot claim that we have discovered and corrected all of the erroneously missing values.

Of the 1,218 USHCN stations, 1,211 have been “completed” with the remaining 7 requiring considerable, yet-to-be-done keying of observations, or are simply too distant from other stations in the early years for reasonable substitution. Three stations in extreme southwestern Texas and the station near Mt Rainer Washington (Longmire) are examples of this latter situation.

It should be noted that the purpose of this dataset is to document the extreme events as observed in individual stations then aggregate to areal means. Each station will have its unique set of issues regarding the impact of non-climatic inhomogeneities on long-term metrics and this should be considered (see also discussion section). There have been numerous attempts to remove non-climatic factors from station data, including those by the author, so as to more carefully identify potential long-term changes or trends in surface temperature (e.g., Alabama: Christy 2002; Christy and McNider 2016; California: Christy et al. 2006; US: Klotzbach et al. 2009; East Africa: Christy et al. 2009; Christy 2013).

The impact on the magnitude of the trend of these corrections is usually less than 0.2 °F (~ 0.1 C) decade^− 1 and for an adjustment to any given year averages much less than 2 °F (~ 1 C). Such is also the case for “adjustments” for non-climatic factors in the USHCN monthly v2.5 dataset (Menne and Williams 2009). So, while the adjustments for these factors are quite important for assessing long-term trends, they have a smaller impact on the magnitude of the much larger extremes of daily temperature which often exceed 30 °F (~ 17 C) from the mean value (e.g., see Fig. 2 above, where the extreme was 40 °F (~ 22 C) above the mean monthly T_Max). Thus, with or without the relatively small corrections for background trend analysis the current dataset can provide useful information about extremes due to their more than order-of-magnitude greater variability relative to the non-climatic influences. Additionally, the availability of 1,211 stations will provide confidence in regional extremes as the pooled data will reduce random error influences.

Many questions may be answered with this completed dataset. To consolidate results we shall ask four specific questions which will be of use for understanding changes through time of these extreme metrics. We begin with a question of wide interest.

4.1 Question 1. When did the all-time record hottest and coldest temperatures occur?

In this metric, all daily T_Max (T_Min) temperatures in May-Sep (Dec-Mar) were examined to discover the year in which the absolute hottest (coldest) single event occurred for each station. Multiple occurrences of an extreme were proportionally distributed so that the sum of all occurrences over the 127 years for a given station is 1.00. While this metric is often sensationalized, it is not very robust as the single hottest (coldest) event can be influenced differentially across a region by many factors during a given year, such as a local rain shower in the midst of an area-wide record heat wave. However, with 1,211 stations, an answer for the CONUS as a whole and the nine large regions should have relatively high confidence.

The results (Fig. 3) indicate that the single years of 1936 (1899) experienced extreme hot (cold) events that dominate the time series, each enveloping about 22% of the CONUS. The next four years in order with large areas for extreme heat were 1934, 2021, 1954 and 1930 all at or below 5%. The four next coldest years were 1912, 1905, 1937, and 1936, also representing less than 5% of the CONUS.

The alternative text for this image may have been generated using AI.

Full size image

Proportion of the CONUS experiencing the All-Time extreme hottest and coldest day by year for 1899–2025

4.2 Question 2. When did the extreme high (low) temperatures occur for each day of the season?

This metric determines for each day of the season the particular year in which the hottest (coldest) T_Max (T_Min) occurred. There are 153 (122) days in the May-Sep (Dec-Mar, leap year) for which a daily record will be achieved. The number of extremes occurring in each year is calculated per station then geographically interpolated as discussed above. This metric is more robust than the single All-Time metric above as each station contributes 153 (122) values to the time series rather than just one. This also provides an indication of the incidence of multiple hot and cold records to help identify periods of excess heat (cold).

The expected value for a purely random process for the number of daily T_Max (T_Min) records would be 1.20 (0.96) in a given year per station for a 127-year record (i.e., 1.20 = 153/127 and 0.96 = 122/127). The results (Fig. 4) indicate again that 1936 contributed the most daily hot records for the CONUS at 6.7 per station but followed more closely by other years, with 1934 (5.3), 1931 (3.4) and 1911 and 1925 (3.3) completing the top five.

The number of coldest records occurred in 1899 (3.7) in association with February 1899 event. The following years experienced extreme cold as well, 1917 (3.3), 1989 (2.9), 1924 (2.4) and 1936 (2.4). Thus, 1936 was a year with many extremes on both ends of the thermometer.

The alternative text for this image may have been generated using AI.

Full size image

Average number of daily T_Max (red) and T_Min (blue) records achieved in each year. Lines represent the 11-year centered average

The 15-year running totals of TMax records of Fig. 4 are given in Fig. 5 for the CONUS and nine subregions (Fig. 1) to show the longer-term variations in daily records. The totals for most of the CONUS rise quickly in the mid-1930s, then decline in the mid-1950s to the lowest point in the 1970’s and 1980’s. Note that the value shown is the “trailing” total, so a time-centered total would be represented by 7.5 years earlier than depicted on the abscissa, indicating that the greatest number of 15-year totals of daily records would be periods centered from 1925 to 1945. From the fewest daily records “centered” around 1970, all regions show some level of rise to the present. The most recent 15-year period for the CONUS experienced 21.1 daily T_Max records, the most since 1936–1950, yet well below the highest of 35.1 in 1925–1939. Dividing 21.1 by 15 (~ 1.41) indicates that the most recent 15-year total was slightly above the expected value of 1.20.

The Pacific SW (33.2), PacNW (25.1) and 4-Corners regions (36.6), all with square symbols, experienced their highest number of daily T_Max records at or near the end of the period. In contrast, regions further east, e.g., Ohio Valley and Upper Midwest (circle symbols), experienced values in the most recent 15-yr period that were over 30 record events fewer than in those 1930s to 1950s. The extreme heat events of 1925, 1930, 1931, 1933, 1934, 1936, 1952–1954 are evident in these running-mean totals, especially for the eastern two thirds of the CONUS. As well, the lack of extreme heat events from 1955 to 1987 for most of the CONUS includes the minimum of 15-yr daily T_Max records in 1965–1979 (8.3 events per station, half the expected value). This feature of extreme heat in the CONUS (primarily in central/eastern regions) during 1925-54 and lack thereof immediately following (1955-87) with only a slow “recovery” (led by heating in western regions) will be evident in all of these extreme T_Max analyses. This result suggests that a general variation in the summer circulation patterns that relates to extreme heating events in the CONUS shifts on a multi-decadal time scale, having favored the eastern regions in the first half of the 20th century and western regions in the 21st century.

The alternative text for this image may have been generated using AI.

Full size image

Fifteen-year running totals (1899–1913 to 2011–2025) of Fig. 4 (hot extremes only) with the CONUS given as gray bars and lines representing the regions identified in Fig. 1

A very different result is depicted in Fig. 6 regarding the temporal distribution of daily record low T_Min. The incidence of daily T_Min records has declined in all regions with the lowest CONUS value of 5.3 occurring in 1998–2012 then ending in 2025 at only 7.9 per station per 15-yrs. This represents a drop of over half from the pre-1997 15-yr values. The relative coherence of the separate regions is also notable with very little variation from the CONUS average in the past 30 years. This suggests that the CONUS-wide response of T_Min is being impacted by a common forcing such as increasing GHGs and/or the ubiquitous surface development around the stations that produces a consistent warming effect on T_Min (see discussion section).

The alternative text for this image may have been generated using AI.

Full size image

As in Fig. 5 except for number of T_Min daily records in 15-yr running-means

4.3 Question 3. How extreme were year-by-year extremes?

The metric chosen here is somewhat different but likely quite useful for the understanding of extremes. For each station, we begin by calculating the hottest (coldest) value for each year. We then define the expected value of the annual hottest (coldest) day as the median of all of the annual hottest (coldest) days in the past 127 years. The departure of the actual yearly hottest (coldest) day from this expected value is then calculated – essentially the anomaly from the median of the expected extreme temperature for each year.

The annual T_Max time series of this value is shown in Fig. 7 where the period from 1925 to 1940 reveals several years with the greatest magnitudes. The early 1950’s also show large areas of excess heat. The early 1990’s tended to be exceptionally cool (in this T_Max metric), likely related to the overall global cooling resulting from the Mt. Pinatubo eruption in 1991 with 1992 being the greatest negative departure. The top five values all occurred prior to 1940 with the most recent “hot” year, 2012, placing 10th overall.

The alternative text for this image may have been generated using AI.

Full size image

Departure (°F) of the single hottest daily T_Max each year from the expected value of the single hottest day. This is the per station average over the CONUS

Regarding departures from the expected coldest T_Min in winter (Fig. 8), the time series is interesting in that its range is over 21 °F (11 C), much larger than that of T_Max (less than 9 °F (5 C)). Note that the scales on Figs. 7 and 8 are different. The event of Feb 1899 is clearly evident as the most extreme event for the CONUS but there are several in the 7–8 °F (~ 4 C) range up until the mid-1980s. From that point, extreme area-averaged cold has almost vanished as only three of the 29 years since 1996 achieved negative departures, all of which were relatively small.

The alternative text for this image may have been generated using AI.

Full size image

As in Fig. 7 for T_Min. Note the vertical axes are different between Figs. 7 and 8

A time series of the difference between 15-year running means of T_Max (Fig. 7) and T_Min (Fig. 8) of the magnitude of the hottest minus the coldest day indicates the CONUS has seen a decline in the ranges of these extremes by approximately 6 °F (3.3 C) in the last 127 years (Fig. 9). This represents a remarkable reduction in the difference between these two extremes prompting the suggestion that extremes are becoming less extreme.

The alternative text for this image may have been generated using AI.

Full size image

Difference of values shown in Figs. 7 and 8 as 15-year running means (1899–1913 to 2011–2025.)

4.4 Question 4. When have heat waves and cold waves occurred since 1899?

While a single daily extreme event is associated with impacts, perhaps a more significant metric is the occurrence of consecutive days of relatively hot or cold temperatures. It is in these events that the cumulative effect of the extremes takes an ever-higher toll as the duration increases (e.g., Chicago 1995 in Chagnon et al. 1996).

There are numerous methods by which one may define heat/cold waves. A common metric is to determine the number of consecutive days that are hotter than (colder than) the 90th (10th ) percentile for heat (cold) waves (USGCRP, 2017). In this metric, percentiles for temperature were calculated for each day using a window of ± 3 days for the 1899–2025 period, giving approximately 850 days available for determining the percentiles for each day. This metric is simply the total number of days per station within consecutive periods (defined below) within which every day exceeded the threshold (hot or cold).

The choice displayed here utilizes a minimum period length of six days and a threshold of 90th percentile for hot days and 10th percentile for cold days. (Event lengths from one to eleven days were examined as well as other percentile thresholds, but all resulted in the same overall features.) The metric here is total days within heat (cold) waves, or the number of days that met or exceeded a percentile threshold and were in periods of at least six days in length.

Figure 10 shows the annual station average number of heat wave days for the CONUS as 15-year running totals of such days (1899–1913 to 2011–2025). The results are in many ways very similar to those of the 15-yr daily record chart above (Fig. 5).

The alternative text for this image may have been generated using AI.

Full size image

Fifteen-year running total (trailing) of heatwave days as defined in the text for May-Sep 1899–1913 to 2011–2025

Of interest in Fig. 10 is the remarkable interregional variety of heatwaves both in terms of the distribution in time and in terms of their magnitude. The four northern-most regions (Pac NW, No East, No Plains, Up Midwest) tend to experience fewer overall heat wave days (25 to 37 days per 15-yrs) as transient summer events are evidently more frequent which shorten periods of extremely hot weather. On the other hand, the southern regions of the CONUS (Pac SW, 4 Corners, So Plains, OH Valley and So East) are apparently more vulnerable to stationary patterns that prolong summer heat events as these five regions range from 53 to 69 days per 15-years. In the most recent 15-yr period (2011–2025) the Pac SW and 4-Corners regions experienced their greatest number of heatwave days with the Pac NW just 2.5 days below its peak in 1960-74. The period 1930–1944 produced the highest number of heatwave days for the CONUS at 84.1 while the fewest occurred in 1965–1979 with a total of 26.1 days per station.

The incidence of cold wave days in the CONUS is characterized by a general decline as seen in the 15-yr running averages in Fig. 11. As will be discussed later, there is substantial evidence that T_Min is affected significantly by even minor human development around the station site over time and this influence is likely of some importance here. Of interest in Fig. 11 is the difference in character of regions that show significant decline (Pac SW, Pac NW, 4-Corners and No Plains) versus regions to their east which have seen little change. Except for the Pac NW, all other regions appear to have converged on a small number of cold wave days in the most recent 30 yrs.

The alternative text for this image may have been generated using AI.

Full size image

As in Fig. 10 except for cold wave days

Comparing the two metrics in Figs. 10 and 11 produces Fig. 12 which displays the sum and the difference, year-by-year of the 15-yr running means. The sum of days in extreme heat/cold declined from over 120 in the 1930s to about 75 since 1960. The conclusion here would be that the CONUS has experienced a decline of around 30% of these durative extreme events in the past 100 years. Along with this decline has been an increase in heatwave days vs. cold wave days since the 1970s, mainly due to the increase in heatwave days in the West (Fig. 10) and the decline in cold wave days overall.

The alternative text for this image may have been generated using AI.

Full size image

The sums and differences of 15-year running means of heat/cold wave days for the CONUS as depicted in Figs. 10 and 11

The evidence from these metrics of extreme events has now been presented and we shall move to a discussion of the caveats and applicability.

We begin the discussion by addressing the impact of non-climatic influences (potential caveats and limitations) on the observations such as (a) sudden inhomogeneities (i.e., changing locations, instrumentation or time of observation), and (b) slow inhomogeneities like urbanization (or general development around the stations), especially in the context of T_Min. From there we shall discuss the overall findings of these results, then provide an example how this dataset can inform readers of more typical results reported to the public.

The literature on the impact of non-climatic influences (NCIs) on surface air temperature measurements is vast and continues to grow, and, as noted, the author has made a number of contributions to this field of study. Given the availability of this literature, which formally began as early as 1833 (London’s urban heat island, Mills 2008), our discussion here will be relatively brief. Readers may want to examine efforts addressing NCIs relevant to stations within the USHCN monthly average dataset as noted in Oke 1973; Karl et al. 1988; Gallo et al. 1999; Peterson 2003; Hausfather et al. 2013; and Katata et al. 2023. A common theme of these studies is that NCIs tend to have much greater impact on T_Min than T_Max. We note that a single station may have ten or more NCI events in which a shift is detected, but that they are generally random in time (Menne et al. 2010). As such, when combining into regional averages, the net impact is very small.

To demonstrate the impact of NCIs on a single USHCN station, we shall draw from the extensive empirical work performed on Fresno California and how it compared with surrounding, more-rural stations (Christy et a. 2006, Christy 2021; Kim and Christy 2022). These studies had the advantage of almost 100,000 additional daily observations for stations within 80 km which had never before been keyed-in to computer-readable files. Fresno is a non-representative example of a typical USHCN station as its population grew from 12,470 people in 1900 to a merged, contiguous metropolitan area with 803,000 in 2024. However, such an example is useful so as to understand the likely upper limit of the influence of NCIs on USHCN stations particularly that of urbanization.

In Fig. 13 we show the bias that is added to the original temperatures for Fresno as produced in Christy 2021 (C21) and the USHCN v2.5 (Menne and Williams 2009 or NCEI) as determined by shifts (sudden NCIs) identified with comparison to surrounding stations. Because C21 incorporated observations that were unavailable to the NCEI process, the neighborhood of comparator-stations was within a closer distance than utilized by NCEI. Note that before this analysis was done, C21 adjusted the pre-1948 data to account for a major shift related to the relocation of the thermometers from a rooftop 30 m above the surface to ground level. In addition, we included new observations from the Hammer Field Army Air Force installation which became the official site for Fresno in 1949, so the adjustments prior to that time are not strictly comparable in this figure. For example, the T_Max biases added by NCEI are substantial prior to 1949, but these were accounted for in C21 as the pre-merged data had already adjusted the T_Max values accordingly. One can see many shifts were identified in both analyses related to relocations or instrument changes which were relatively frequent as the city and airport grew.

The alternative text for this image may have been generated using AI.

Full size image

(Top) Bias to be added to the raw T_Max monthly values for Fresno California as determined by Christy 2021 (C21) and Menne and Williams (2009; updated, NCEI). (Bottom) as at top except for T_Min

Once the shifts were accommodated, the time series (Fig. 14) for Fresno 12-month running anomalies indicate very different results between T_Max and T_Min, which is a clear indication of the NCI of urbanization. The T_Max time series indicates no trend through 2012 (slightly negative) but contains a relatively sudden rise in 2013 which is consistent with the entire western CONUS as seen in Figs. 5 and 10. The overall T_Max trend is + 0.03 °F (+ 0.02 C) decade^− 1. The trend in T_Min is + 0.43 °F (+ 0.24 C) decade^− 1.

The alternative text for this image may have been generated using AI.

Full size image

Time series of 12-month running mean of monthly TMax (red) and TMin (blue) anomalies for Fresno CA after being adjusted to remove sudden NCIs

The steady rise in T_Min values relative to T_Max suggests a cause that is consistent with urbanization as noted in numerous publications (e.g., Oke 1973). To give an idea of the magnitude of this urbanization effect, in Fig. 15 we show the decadal progression of these temperature anomalies relative to a base period of the first 50 years (1905–1954) of Fresno vs. the nearby comparator stations (total of 28, some being USHCN as well), all of which are much smaller in population, and in a few cases, still rural. The comparators were also adjusted for their shifts.

The alternative text for this image may have been generated using AI.

Full size image

Time series of decadal anomalies relative to 1905–1954 in T_Max (red Fresno, orange comparator stations) and T_Min (blue light circles Fresno, blue solid circles comparators)

The clearest result here is the significant warming of Fresno T_Min relative to the comparators of over 5 °F (2.8 C) in the past 80 years (blue lines). Such an effect will certainly impact the T_Min results shown in figures of T_Min metrics for Fresno and any other station with major urbanization issues. It remains for on-going research to calculate the impact, station-by-station, for the entire dataset however. As noted above, we assume that Fresno is one of the extreme cases of NCIs and that the CONUS-averaged results for T_MIn presented in the previous section would be minimally contaminated by this type of NCI, though with the substantial growth in the Pac SW and 4-Corners regions we might expect an area-averaged NCI signal there. A likely result with urbanization removed would be a small increase in cold events in the last few decades (and a lessening in the first few), the change being small partly due to the lack of wide-spread, extreme cold air events since 1990.

A more encouraging result is evident for T_Max. Here, Fresno’s urbanization impact appears to be less than 2 °F (1 C). The mechanisms for producing this more modest impact are well-documented as summer daytime heating generates vertical and horizontal motions that mix the deeper atmosphere down to the surface, negating much of the impact of the extra heating caused by impervious surfaces such as roads, parking lots, buildings, etc. (e.g., Spencer et al. 2025a). (For a descriptive summary of these processes see Christy et al. 2009 and for a modeling analysis of boundary layer meteorology see McNider et al. 2012.) Thus, at least for the CONUS as a whole, there is likely an insignificant, though spuriously positive, impact of urbanization on T_Max metrics.

Irrigation has generally increased across the CONUS in the past 100 years affecting local temperatures (Christy et al. 2006). Mueller et al. (2016) identified areas in the CONUS that indicated a temperature response to irrigation. To investigate this impact on our results, we eliminated stations in those areas with a significant irrigation signal (about 2% of the CONUS) and recalculated the extreme metrics. The difference was minuscule, being about 0.2% per century relative to the full CONUS for heatwave days.

Though not as relevant in the more sparsely occupied western and northern CONUS, the greater presence of aerosols during the 1960’s to 1980’s may have suppressed T_Max values (and increased T_Min values) in portions of the eastern CONUS (Bauer et al. 2020). Leibensperger et al. 2012 estimated a cooling in the most-affected area of the mid-eastern CONUS of approximately 1 °F (0.5 C) during the period of highest emissions (1965–1985). An experiment was conducted in which 1 °F was added to T_Max values in this section of the CONUS for 1965–1985 to understand the scale of any impact of this potential non-climatic cooling. The result with reference to Fig. 10 (heatwave days) for example, indicated decreases for the non-aerosol period of 3% (i.e., almost imperceptible) and increases in the aerosol period of an average of 14% from the base average. Because this aerosol period was somewhat near the middle of the overall time period, trends were not affected. As well, the 1965–1985 period was already characterized by fewer heatwave days, so that increases of 14% upon already small values did not alter the overall result. Thus, an artificial modification of the observations to incorporate a possible aerosol impact does not change our fundamental results when considering the period beginning in 1899.

Finally, we note here that the NCI of “time-of-observation” bias (T_Obs) has no impact on questions 1 to 3 (the single annual or period-of-record extreme is not a function of the time of day when it is recorded) but may on Question 4. To test this, we manually examined the original observational forms to discover if the T_Max recorded in the late afternoon may have been incorrect and thus allowed a heat wave to be extended longer than in actuality. We looked for those cases in which the T_Obs temperature (late afternoon) was sufficiently below that of the recorded 24-hour T_Max value which then likely represented the heat from the day before when the thermometer was read and re-set.

For a category of T_Max being above the 90th percentile we found 0.9% of events may have been erroneously counted because the calendar day T_Max was very likely below the 90th percentile although the recorded 24-hour T_Max (late afternoon to late afternoon) was above it. In the most significant case, an 11-day heat wave was found to have its 6th day to likely have been below the 90th percentile threshold, thus breaking the stretch into two 5-day episodes each of which failed to meet the 6-day minimum, thus removing 11 heatwave days from the time series. Accounting for this we determined that the incidence of 90th percentile heat-wave days was about 4% too high in the early part of the record as compared with the present (Fig. 10). This indicates the artifact did not alter the results in a significant way.

With these NCI issues discussed we now turn to the meaning of the results we have calculated.

Overall, our project indicates that extremes in summer heat-related metrics for the CONUS as defined in the four questions above do not show increasing trends, but rather modest negative trends, and thus appear to be substantially affected by other forcings such as natural variability in addition to GHGs. There are positive T_Max metric trends in western regions which are countered by larger negative trends elsewhere.

The number of cold extreme events has declined in the past 30 years too and this is likely, in part, related to the development of infrastructure around the stations which disturbs the nocturnal boundary layer, inhibiting the formation of the cold, shallow layer in which T_Min is observed. Additionally, this result may be an early sign of atmospheric warming of the coldest air masses by the added GHGs (e.g., Krayenhoff et al. 2018), though this hypothesis has not been confirmed as a direct result of GHGs (e.g., Huang et al. 2023). Observations of the deep atmospheric temperature in the polar region north of + 60° latitude indicate a warming trend of + 0.47 °F (+ 0.26 C) decade^− 1 since 1979 compared with a global trend of + 0.27 °F (+ 0.15 C) decade^− 1 (Spencer et al. 2017). This would suggest Arctic air intrusions into the CONUS may be slightly warmer now than in the past century or so (for whatever reason) and thus consistent with the results shown here for a lessening of the magnitude of cold events in recent decades. However, we note the same area in the southern hemisphere shows virtually no warming (+ 0.05 °F (+ 0.03 C) decade^− 1).

These T_Max and T_Min results suggest that the magnitude of local and regional short-term natural variability in the CONUS, especially for T_Max, is greater than the magnitude of warming due to GHGs and that a signal tied to GHGs has yet to emerge in these metrics. This does not imply that there is no detectable signal of GHG forcing in the global climate system (e.g., Spencer and Christy 2023), but that at local and regional levels, the magnitude of day-to-day weather variations of the type examined here essentially swamps the background GHG signal in the CONUS. The climate system is composed of two turbulent fluids (atmosphere and ocean) whose interaction leads to an infinite variety of local realizations in the CONUS which include the as-yet-unmatched Feb 1899 cold episode and the heatwaves of the 1930’s.

As stated in the introduction, we seek to analyze the station data to answer questions about changes over time of these extremes and thus take a step toward understanding how increasing GHGs are related. There is an enormous number of claims regarding the relation between weather extremes and GHG concentrations, but will shall examine only one source. It is important to note that we are testing what might be called “headline” claims from a government report that make their way through the information sphere and that we simply desire to demonstrate how this dataset may be applied to test claims.

We shall investigate a few claims about changing hot temperature extremes from the National Climate Assessment Climate 5 (NCA5, Jay et al. 2023) which states in many places that “Climate Change” i.e., change induced by human emissions of GHGs, is causing observed changes. For example, this statement is fairly clear and bold,

• The evidence for warming across multiple aspects of the Earth system is incontrovertible, and the science is unequivocal that increases in atmospheric greenhouse gases are driving many observed trends and changes. (Ch 2, Introduction).

NCA5’s claim connects a rise in global surface air temperature to GHGs and then with changes in numerous types of non-specified extreme weather events. Later, and related to our study here, NCA5 states regarding the US, “Climate change is increasing the frequency and severity of many extreme weather and climate events, including heatwaves …” (Jay et al. 2023; ch 2). One of the reasons for our effort here is to provide observations to quantify claims such as this.

When tracking down this “increasing heatwave” claim (recall that NCA5 connects this claim’s cause to increases in GHGs) we are referred elsewhere to a statement with more clarity and nuance, explaining that, “Since the 1960s, the frequency of heatwaves … [has] steadily increased in certain areas (NCA5 App 4 Sect.  2).” This elaboration is more equivocal as it identifies a specific period (“since 1960”) and an undefined but specific region (“certain areas”). Our results (from Fig. 10) indicate an insignificant positive CONUS trend in heatwave days (3% since 1960), though specific regions in the Southwest quadrant have seen increases of up to 10%. The other regions indicate trends being zero or negative, and, beginning in any year prior to 1937, the full CONUS trend is always negative. Thus, the statement by NCA5 of increasing heatwaves since 1960 is confirmed in terms of sign, but not in terms of confidence, i.e., being insignificant.

NCA5 offers more specific information in the appendix that allows for testing:

• Several major heatwaves have affected the US since 2018, including a record-shattering event in the Pacific Northwest in 2021. The western US has been particularly affected by extreme heat since the 1980s (Fig. 2), experiencing a larger increase in days over 95 °F, as would be expected given the greater warming in that region relative to the eastern US. By contrast, the number of very hot days has actually decreased across the central and eastern regions due to summer cooling trends in the region. (NCA5, App 4, Sect.  2).

Here we have even more detailed information to let us know that (a) in 2021 a serious heatwave occurred in the Pacific Northwest, (b) the western CONUS has experienced more extreme heat since 1980 including more 95 °F days than most of the CONUS, (c) that this is to be expected, and (d) that the eastern US has experienced decreasing extreme heat-related metrics.

The 2021 Pacific Northwest heatwave was examined in detail in Mass et al. 2024; indicating that this historically unprecedented extreme event would have happened in any case due to a circulation pattern that was actually inconsistent with the type of circulation patterns inferred by climate models with enhanced GHG forcing. However, our results above confirm that the western US has experienced more extreme heatwaves since 1980 (Fig. 10), though, the use of the threshold of 95 °F days may be criticized for two reasons.

First, because this is a nonlinear (i.e., threshold) metric, stations which are located in areas with few such days will contribute little to the overall understanding, thus this metric is driven by a subset of stations which have large variability around the threshold value (one of the reasons this metric was not part of our four questions). As seen in Fig. 16, regions in the West and Southwest dominate the metric and thus the results (dashed lines indicate regions with positive trends). Secondly, trying to relate this to increasing GHGs is difficult, as this metric should be changing throughout the CONUS if indeed GHGs are its fundamental driver.

As shown in Fig. 16, there is no increase in the occurrence of 95 °F (35 C) days within the CONUS, and in fact, the CONUS has experienced an 8.3% decline since 1899. None of the past 10-year totals are in the top ten values. The NCA5 acknowledges the decline for the central and eastern parts of the CONUS (see their Fig. 2) yet ascribes this decline to an apparently untestable cause, “summer cooling trends.” To be consistent, the NCA5 could also report that the rise in western CONUS extreme temperatures is due to “summer warming trends”.

The alternative text for this image may have been generated using AI.

Full size image

Average number of days per station equal to or above 95 °F (35 C) for the CONUS (bars) and regions (lines, 11-year centered averages). Dashed lines indicate regions with positive trends

Calculating the percentile exceedances provides a more informational metric. In this case, every station contributes equally based on its own heat-related metrics so that one or two regions do not dominate the areal results. Figure 17 displays the number of days a station equals or exceeds its 95th percentile value. The result is much more coherent among the regions and demonstrates the CONUS as a whole has many common features such as heat in the 1930’s, 1950’s and cool years in the 1960’s and 1970’s. The impact of the 1930s heatwaves is clearer since relatively cool regions (e.g., Northeast) have few absolute threshold temperature exceedances (Fig. 16) but many percentile threshold exceedances (Fig. 17).

The alternative text for this image may have been generated using AI.

Full size image

Average number of days per station equal to or above the 95th percentile for the CONUS (bars) and regions (lines, 11-year centered averages)

The extreme heat in the pre-1940s era provides a measure of the magnitude that natural variability exerts on the heat-extreme climate of the CONUS. Such eras increase the statistical magnitude of natural variability making it difficult to attribute recent events to a specific cause, realizing that natural variability is always present in any climate situation. The overall picture from results such as seen in Figs. 5, 6, 7, 8, 9 and 10 is that the heat extremes occurring in the CONUS today are well-within the range that natural variability already provides. However, we note that the Intergovernmental Panel on Climate Change has examined heat extremes (generally only since 1950) and demonstrates that these are increasing in frequency and attribute this result primarily to rising GHG concentrations (e.g., see Seneviratne et al. 2021).

We acknowledge too that for a given array of stations experiencing a gradual warming trend, there should be a gradual increase in various extreme heat metrics as noted in Seneviratne et al. 2021 for the globe related to enhanced concentrations of GHGs. Indeed, our recent calculations of total heat content in the oceans and land masses indicates an increase over the period of 1970–2021 associated with rising temperatures in those bodies which we presume is due to the enhanced greenhouse effect (Spencer and Christy 2023). However, as noted above for the CONUS, the extreme heat of the pre-1940s still dominates these regional results.

In the field of climate change, attention has been drawn to extreme metrics occurring in the last several years as evidence for human influences through increasing GHGs (e.g., USGCRP 2017; Seneviratne et al. 2021; Jay et al. 2023). Examining this aspect of climate and weather is appropriate since human thriveability is often constrained by the magnitude of the extremes that we experience. We describe here a dataset constructed to examine the occurrence through time of extreme temperature metrics in the CONUS for the coldest winter and hottest summer days since Dec 1898. The dataset is based on the 1,218 USHCN stations 1,211 of which have been supplemented to be “complete”, i.e., each station having at least 92% of days available for analysis. The results indicate that extremes in heat-related metrics for daily T_Max in the summer have not increased and in fact often show modest declines since 1899, due mostly to the early heat events during 1925–1954. This is consistent with Seneviratne et al. 2021 (IPCC AR6, their Fig. 11). Cold-related extreme events based on winter T_Min show evidence of decreasing occurrences, two causes of which were proposed, (1) increasing human development around weather stations, and (2) an early response to increasing GHGs as they warm the coldest air first. When taken together, the occurrences of heat and cold extremes have declined over the past 127 years in the CONUS, i.e., the climate over the CONUS has become less impacted by temperature extremes to this point. Relating this reduction of extreme events to increasing GHGs would be difficult as the magnitude of the regional natural variability of weather and climate is considerable in comparison to a small GHG-induced temperature rise.

The impact of Non-Climatic Influences was considered in the temperature evolution of one USHCN station, Fresno California, as an example of a clear and large response to forcings unrelated to the increasing GHGs. In this case, the urbanization impact on T_Min of 5 °F (~ 3 °C) is clearly apparent, while summer T_Max (with urbanization) indicates a trend not significantly different from zero. Voluminous research has and will be performed on this aspect of surface temperature records as these types of influences need to be identified and removed so that changes in the background climate due to GHGs may be estimated with more confidence. We also demonstrated that one must be cautious when interpreting official statements about extreme weather events for the CONUS.