Methodology

Everything on this site is computed from public NOAA data with a method simple enough to state in one line — and audited carefully enough that we’ll state its limitations too.

The formula

Climate-Adjusted Normal(month) = Official normal(month) + rate(month) × 2.05 decades

The level is NOAA’s 1991–2020 normal for the county, computed from the homogenized nClimDiv record. Because a 30-year average best represents its midpoint, it is treated as an estimate of the climate of mid-2005.

The rate is the county’s warming trend in °F per decade, estimated separately for each calendar month with a hinge fit: temperature is modeled as flat before 1975 and linear after, following the approach developed in NOAA research (Livezey et al.) to avoid understating the modern trend by averaging it with the mid-century plateau. Fitting each month separately preserves the seasonal fingerprint of warming — in most counties winter is warming roughly twice as fast as summer.

The advance moves the normal 20.5 years, from mid-2005 to 2026.

Why homogenized county data, not raw station readings

Thermometer records contain artifacts: stations move, instruments are replaced, observation times change. In one California county we audited, an automated-instrument changeover in the 1990s put a false ~2.5°F step in the raw overnight-low record — enough to swing a naively fitted trend from negative to strongly positive depending on the start year chosen. NOAA’s nClimDiv dataset corrects these artifacts; trends fit to it are stable to within a few hundredths of a degree per decade regardless of start year. That stability is the entire reason this site can exist.

Data sources

• NOAA nClimDiv county monthly temperature series, 1895–present (homogenized; the trend source, and the level source at county resolution).
• NOAA U.S. Climate Normals, 1991–2020 (the official-normal definition this site advances).
• U.S. Census county names and identifiers.

Limitations, stated plainly

• The Climate-Adjusted Normal is a projection — a fitted trend read at the present day — not an observation. It carries statistical uncertainty of roughly a few tenths of a degree, larger in counties with noisier records.
• Day-to-day weather dwarfs the adjustment. Daily temperatures vary around any normal by several degrees; the adjustment shifts the yardstick, not the weather.
• County resolution smooths local detail. Elevation and coastal microclimates within a county are not resolved. The seasonal shape shown on county pages is smoothed from monthly values.
• Alaska coverage begins in 1925 in the NOAA county record; Hawaii and the territories are not yet in nClimDiv and are not yet on this site.
• Linear extrapolation is conservative. If warming accelerates, the Climate-Adjusted Normal will still lag reality — just by less than the official one.

Validation

For our audit county (San Joaquin County, California), three independent estimates of the gap agreed within a third of a degree: the hinge fit (+1.6°F), a simple linear trend from 1948 (+1.3°F), and a no-model observation — the shift between the 2011–2025 average and the official normal (+1.3°F). As a further check requiring no model at all: if the official normal were current, about half of days would exceed it; in the last decade, 60.7% did.

Reproducibility

Every input is free, public NOAA data, and the method above is complete — a competent analyst can reproduce any number on this site. We consider that a feature.