What the data shows
Four things stand out.
Variability does the sorting
Variability tracks the final score (r = 0.85) more than comfort does (0.67). 315 places get a perfect 100 on it. Jordan, Montana gets 2.5. Places sink when the weather keeps swinging, not just when it is cold.
Comfort and health disagree
The two are slightly negatively correlated (r = -0.15). Brownsville and Key West score near 90 on Health and in the mid-30s on Comfort: hot and humid, almost never cold. So I publish both.
Geography is lopsided
California averages 80.8 raw across 116 stations, the best of any state with at least eight. North Dakota averages 57.2 across 29, the worst. Washington is second at 80.0. 78 of the worst 100 places are in North Dakota, South Dakota, Montana, Nebraska, Wyoming or Minnesota.
Rain barely matters
Rain is 10 percent of the score because it costs outdoor time, not mood.[7] Brookings, Oregon gets 119 inches a year, the wettest in the data, and still ranks 48th. Mild, steady weather outweighs a lot of rain.
Four calls I had to make
The research does not settle everything. Here is where I had to choose.
Which is worse, heat or cold?
It depends what you ask. For comfort, heat. People pay more to avoid hot days than cold ones,[3] and outdoor work falls off sharply above 85°F.[7] For health, cold. It causes 7.29% of deaths against 0.42% for heat, mostly from ordinary chilly days, not extremes.[1] So Comfort counts heat 1.5 times as much as cold, and Health leans the other way.
Big swings cost something
Wide daily temperature swings raise heart-related deaths about 0.7% per °C.[6] Freeze-thaw wear is worst where temperatures keep crossing 32°F. Variability is 15% of the score. I set no cutoff, because no study gives one.
Does snow count as a perk?
Skiing is a $59bn industry, but no study has put a value on snow itself.[9] So I gave it a small bonus: at most 5 points, only above 7,000 feet, enough to soften mild cold but never harsh cold. 32 towns qualify. It is a guess.
Are continental climates extra bad?
I expected hot summers plus cold winters to hurt more than each does alone. No study backs that, so I added no extra penalty. St. Louis still does badly the plain way: humid summers, cold winters, big swings. It ranks 1,100th of 1,831 (raw score 70.4), mid-pack.
Tap a card for the breakdown. Eagle is the airport serving Vail, which has no station.
The best and the worst
The top ten is all Southern California coast. The bottom is the Northern Plains, where variability does what cold alone does not.
Best ten
Top of the listWorst ten
Bottom of the listSuperlatives
The extremes in 13.4 million days of station data. Tap a card for the full record.
The full leaderboard
Search a place, sort, filter by badge, or tap + on two places to compare. The big number is the rescaled score. The raw score sits beside it.
How the score works
The weights are a choice, not an estimate.
Comfort (50%), weighted toward heat
Built from perceived temperature: NWS heat index, wind chill and shade-UTCI. Each day takes the worse of the heat-index and UTCI readings. Heat counts 1.5 times cold, around a 65°F optimum.[3][4][5]
Health (20%), weighted toward cold
A continuous penalty for distance from the roughly 74°F minimum-mortality temperature, in both directions, with cold 1.25 times heat. Most temperature deaths are moderate cold, so there is no cliff below 20°F.[1][2] The 1.25 is a judgment. The 9-to-17-to-1 ratio of cold to heat deaths mixes how often cold happens with how bad it is.
Duration lives inside heat and cold
Heat waves (three or more days at or above the place's 92.5th percentile, or a heat index of 103°F) and cold spells (two or more days at or below its 5th percentile, or a mean of 20°F or less) are multiplied by day: ×1.0, ×1.1, then ×1.25. That matches the measured +2.8% a day for heat and +1.5% for cold.[8] Warm nights in heat waves show in each place's detail view.
Variability (15%), continuous
Five annual measures: temperature SD, day-to-day change, diurnal range, seasonal amplitude and freeze-thaw days. Each is scored against the national distribution. The median costs nothing. The 90th percentile costs full marks.
Rain (10%), frequency not totals
I count wet days: drizzle 0.2, wet days 0.5, heavy days 1.0. Any rain deters time outdoors.[7] A wet day costs 1.5 times more if it would have been comfortable, and 1.5 times more in the driest quarter. That is why Seattle's drizzle costs so little. I price rain as lost outdoor time, not sadness.
Extreme (5%), a weak proxy
Days with over one inch of precipitation or mean wind over 25 knots. Daily data cannot identify hurricanes, so this is small on purpose. 375 places score 99 or higher.
The rescaling
Raw scores run from 53.4 (Jordan, MT) to 94.3 (Carlsbad, CA), median 72.3. I stretch that to 0 to 100 so differences are readable. It is relative: a 50 does not mean half of perfect weather. Rank order is the same either way.
A problem in the source data
The dew-point column is mislabelled. It is Celsius, not Fahrenheit. As Celsius it matches the file's humidity column (r = 0.977). As Fahrenheit it does not (0.669). I convert first. It matters most in humid places: Miami's Comfort score is 46.2 and its Health score is 92.0.
What the data cannot do
- Shade only. No sunshine or cloud data, so UTCI assumes shade. Full sun adds up to 15°F to the heat index and 10 to 18°F to wind chill, so sunny high-altitude places are understated.
- Snow is a proxy. No snowfall data. A snow day is 0.01 inch or more of precipitation on a day averaging 32°F or less.
- Daily, not hourly. Minneapolis's 2019 polar-vortex wind chill computes to -51°F against the NWS's -56°F.
- No acclimatisation. A 90°F day in Phoenix is not a 90°F day in Seattle. City-relative percentiles help for heat waves and cold spells. The absolute penalties do not.
- No hurricanes. See Extreme.
- Weights are a choice. The 50/20/15/10/5 split is a starting point for testing, not something any study measures. Sub-scores are published for every place.
Method and fine print
Source: Dewey Data "Custom Weather, Daily Weather Data" (api.deweydata.io), 13,440,992 daily observations. Each place is its best-covered weather station. Stations with fewer than 2,500 days between 2018 and 2025 were dropped, leaving 1,831 places in 49 states. The formulas match NWS charts (heat index at 90°F and 70% humidity is 106°F, wind chill at 0°F and 15 mph is -19°F) and real extremes (Phoenix, July 19, 2023: 117°F. Miami, July 15, 2023: 104°F).
Excluded: placeholder rows dated 1970, five Gulf of Mexico oil platforms, 35 stations whose rain gauges read over 2.5 times their neighbours (five genuinely soaking ones were kept), and 31 station codes I could not match to a place. A "place" is where a station sits. Most are towns and cities, plus a few airfields, a naval air station (San Clemente Island, 10th) and mountain passes. The weather there is real. Impossible hottest or coldest values (a -79.6°F low in Alabama) show as not reported. Scores are unaffected.
References
- Gasparrini A, et al. Mortality risk attributable to high and low ambient temperature: a multicountry observational study. The Lancet 2015;386:369-75. Cold 7.29% vs heat 0.42% of deaths across 74,225,200 deaths in 384 locations; most of the burden is moderate cold (6.66%).
- Zhao Q, et al. Global, regional, and national burden of mortality associated with non-optimal ambient temperatures, 2000-2019. Lancet Planetary Health 2021;5:e415-25. Cold 8.52% vs heat 0.91%, roughly 4.6M vs 0.49M deaths per year.
- Albouy D, Graf W, Kellogg R, Wolff H. Climate amenities, climate change, and American quality of life. Journal of the Association of Environmental and Resource Economists 2016;3(1):205-46. Americans most prefer a mean daily temperature near 65°F and pay more on the margin to avoid excess heat than excess cold.
- Jendritzky G, de Dear R, Havenith G. UTCI, why another thermal index? International Journal of Biometeorology 2012;56:421-8; and Bröde P, et al. Deriving the operational procedure for UTCI. Int J Biometeorol 2012;56:481-94. "No thermal stress" is 9 to 26°C (48 to 79°F).
- National Weather Service. Heat index (Rothfusz regression; Danger at 103 to 124°F, Extreme Danger at 125°F and above) and wind chill chart and FAQs (2001 formula). weather.gov.
- Weidmann MD, et al. The association between temperature variability, morbidity and mortality: a systematic review and meta-analysis. Hygiene and Environmental Health Advances 2025;14:100123. Diurnal range: cardiovascular mortality +0.7% per °C, respiratory morbidity +1.0% per °C.
- Graff Zivin J, Neidell M. Temperature and the allocation of time. Journal of Labor Economics 2014;32(1):1-26; Tudor-Locke C, et al. 2006; Tucker P, Gilliland J. Public Health 2007;121:909-22.
- Chen J, et al. Heat and cold waves and cardiovascular mortality in 31 Chinese cities, 2007-13. Environmental Research Letters. Each extra day adds +2.84% (heat) and +1.52% (cold).
- National Ski Areas Association: 61.5M U.S. skier visits in 2024-25 across 492 areas; U.S. ski economic activity about $59bn (2023). No hedonic or livability study valuing snowfall itself was found.