Heat Waves

Four heat metrics from GHCN-Daily stations and Berkeley Earth gridded land data, for CONUS and Northern Hemisphere land

Dataset, region and window

Dataset

Region

Variables on this page

Variable	Finding

Detection

A change is called detected when the Mann–Kendall test finds the trend unlikely to have arisen by chance against a background of internal variability:

Likelihood

Mann–Kendall two-sided p < 0.10. The reported rate is the Theil–Sen slope.

Recomputed per window

The test runs on whatever window you select, so narrowing the window changes both the slope and the verdict.

Per dataset and region

Each combination is tested on its own and the verdict describes only the series on screen. GHCN and Berkeley Earth can reach different verdicts for the same variable; switch between them to see it.

Each chart also draws the 66% and 90% historical range bands so the fitted change can be compared with ordinary year-to-year spread by eye — a trend can clear p < 0.10 while remaining small against that spread. Bear in mind too that a linear trend is a poor summary of a series whose largest feature is a single decade; for CONUS the 1930s dominate.

Methodology

Every metric runs through one implementation (scripts/metrics.py) for all four dataset and region combinations, so a difference between two panels is a difference in the data rather than in the arithmetic. A site is a station for GHCN and a land grid point for Berkeley Earth.

Data and reproducibility

Code is available on GitHub.