A new peer-reviewed paper by John R. Christy (University of Alabama in Huntsville, retired Alabama State Climatologist) examines daily temperature extremes across the contiguous United States (CONUS) from 1899 to 2025.

Published in Theoretical and Applied Climatology, it analyzes over 40 million raw daily observations from an expanded and gap-filled U.S. Historical Climatology Network (USHCN) dataset of 1,211 stations (97.6% coverage, each with at least 92% data completeness). No homogenization, spatial interpolation, or temporal adjustments were applied to the station-level data—only inverse-distance weighting on a 0.5° grid to aggregate while preserving local signals.

This is a peer-reviewed paper by John R. Christy (University of Alabama in Huntsville, retired Alabama State Climatologist), published in Theoretical and Applied Climatology (2026, open access).

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

Springer Nature Link

Published: 18 April 2026

Volume 157, article number 309, (2026)

DOI: 10.1007/s00704-026-06200-3

By John R. Christy

Abstract

Knowledge of temperature extremes, and their potential changes within a climate system of increasing greenhouse gases, is of vital interest for humans and the infrastructure which supports them. To produce a better understanding of how daily extreme temperatures have changed over time in the conterminous US (CONUS), the United States Historical Climate Network (USHCN) database was extended back to 1899 and forward to 2025. The original 1,218 stations, selected in the 1980s by NOAA as capable of addressing climate concerns, have since been neglected – almost half of the stations have closed since 2000. Incomplete station records were supplemented with nearby stations with high correlation and removeable biases to provide time series for 1,211 of the stations with at least 92% of data present. Extreme temperature metrics for summer daily maximum temperatures and winter daily minimum temperatures were calculated. The general result is that metrics for extreme summer heat, e.g., hottest values, number of heatwave days, etc., show modest negative trends since 1899. Extreme cold temperature metrics also indicate a decline in their occurrences especially since the 1990s. In sum, instances of both hot and cold extreme metrics have declined since 1899. To demonstrate an application of this dataset we examined the claims of one source regarding changing temperature extremes, The National Climate Assessment 5.

Data and Methods

Christy extended the U.S. Historical Climatology Network (USHCN) dataset—originally ~1,218 stations selected by NOAA in the 1980s for climate monitoring—to cover December 1898–March 1899 through summer 2025. Nearly half the stations have closed since 2000, so he supplemented incomplete records by “threading” data from highly correlated nearby stations (median correlations ~0.93 for daily Tmax, ~0.89 for Tmin), with small biases adjusted. This yielded 1,211 stations with at least 92% data completeness (median 98%).

• Raw observations only: Over 40 million daily values. No homogenization, spatial interpolation, or other adjustments beyond gap-filling and time-of-observation bias handling via lag correlations.
• Aggregation: Station-level extreme metrics gridded at 0.5° × 0.5° using inverse-distance (1/d²) weighting, covering 97.6% of the contiguous U.S. (CONUS).
• Seasons analyzed: Summer heat via daily maximum temperatures (Tmax, May–September); winter cold via daily minimum temperatures (Tmin, December–March).
• Extreme metrics:
  • Annual hottest/coldest day per station.
  • Frequency of daily records (new highs/lows for each calendar day in the season).
  • Departures of annual extremes from the long-term median expected value.
  • Heat/cold waves: Runs of at least 6 consecutive days above the local 90th percentile (Tmax) or below the 10th percentile (Tmin), using a ±3-day window over the full record for percentiles (~850 days per window).
  • Additional checks: Days ≥95°F, etc.

This approach prioritizes preserving local station signals over heavily processed gridded products.

Key Results on Heat Extremes (Tmax)

Metrics for extreme summer heat show modest negative (declining) trends over the full 127-year period:

• Hottest annual Tmax values, frequency of daily heat records, and heatwave days all trend modestly downward.
• Strong clustering of intense heat in 1925–1954, especially the 1930s Dust Bowl era. 1936 stands out: ~22% of CONUS stations set their all-time hottest day; ~6.7 daily Tmax records per station on average.
• 15-year running totals of daily Tmax records: Peak ~35.1 (1925–1939), low ~8.3 (1965–1979), recent (2011–2025) ~21.1 (still below early peaks and near or slightly above the long-term expected rate of ~1.2 per station per season in some windows).
• Heatwave days (≥6 consecutive above 90th percentile): CONUS peak ~84 per station (1930–1944), low ~26 (1965–1979). Recent periods remain well below 1930s levels; overall sum of extreme (hot + cold) wave days declined ~30% since the early 20th century.
• Days ≥95°F: ~8.3% decline CONUS-wide since 1899. Recent totals not in the top historical ranks. newswise.com

Regional nuance: Some Western areas (e.g., Pacific Southwest, Four Corners) show recent upticks in heatwave activity (up to ~10% since 1960), while central/eastern regions (e.g., Ohio Valley, Upper Midwest) show stronger declines. Nationally, no clear long-term increase.

Key Results on Cold Extremes (Tmin)

Extreme cold metrics show clearer declines, accelerating since the 1990s:

• Fewer record low Tmin days and less severe cold waves.
• 1936 and especially 1899 (Great Arctic Outbreak) had prominent cold extremes; post-1990s, very few large negative departures from expected coldest days.
• Cold wave days (≥6 consecutive below 10th percentile) declined overall, more so in western regions.
• The annual range between the hottest Tmax and coldest Tmin (via their departures) narrowed by 6°F (3.3°C) over 127 years, suggesting reduced temperature variability (“extremes becoming less extreme”).

Discussion, Attribution, and Comparison to Other Assessments

• Natural variability dominates: Early 20th-century heat (tied to phenomena like the Dust Bowl) sets a high bar that recent decades have not consistently exceeded. Regional weather patterns often outweigh any modest GHG signal in these tail metrics.
• Possible influences on cold declines: Urbanization/land-use changes around stations (disproportionately warming nighttime Tmin; e.g., >5°F effect noted in Fresno examples) and, secondarily, an early GHG response that preferentially warms the coldest air masses.
• Urban heat island and other non-climatic effects: Acknowledged and quantified where possible (generally small net impact on Tmax or CONUS averages after processing; minimal on heat metrics).
• Comparison to NCA5: The paper tests claims in the U.S. National Climate Assessment 5 (e.g., increased heatwave frequency/severity since the 1960s, especially in the West). Findings show a small positive CONUS heatwave trend since 1960 (~3%, statistically insignificant), confirming the sign in some regions but not the magnitude or confidence when viewed over the full record. Pre-1960 extremes are downplayed in shorter-baseline analyses.
• Consistency with broader literature: Aligns with IPCC AR6 (limited changes in some heat metrics globally) but emphasizes the value of raw, long-record station data over adjusted products or selective periods.

Conclusions from the Paper

“In sum, instances of both hot and cold extreme metrics have declined since 1899… the climate over the CONUS has become less impacted by temperature extremes to this point.” Relating the overall reduction directly to rising GHGs is difficult, as “the magnitude of the regional natural variability of weather and climate is considerable in comparison to a small GHG-induced temperature rise.”

Christy describes the work as “a labor of love and curiosity” to better compare today’s events against the full historical context using actual observations.

Context and Limitations

This analysis does not deny modest overall warming or regional changes (e.g., Western heat upticks). It highlights that U.S. temperature extremes—the high-impact tails—have moderated rather than amplified over 127 years when using a comprehensive, minimally processed dataset. Shorter periods, absolute thresholds (vs. percentiles), or heavily adjusted data can emphasize different signals. Station network degradation and local siting/urban effects introduce uncertainties, particularly for Tmin.

The full paper is available via Springer (open access). It contributes to ongoing debates on extremes vs. means, data processing choices, and the relative roles of anthropogenic trends versus natural variability. Science advances by rigorously testing assumptions against long-term observations, and this provides one such detailed test focused on daily CONUS station data.