How the analysis works

Each check removes the seasonal cycle and measures the change per decade, with a 95% range that allows for one month's value being tied to the next (Newey–West). A second, rank-based test (seasonal Mann–Kendall) and a robust slope (Theil–Sen) must agree before a trend counts as a “Yes”; otherwise it is “Not proven”. Windows that span NASA's instrument change around 2000, or are short, carry a warning.

A test needs at least 10 years, 120 months with data, 90% of months present and 8 of each calendar month. Missing months are never filled in. Gaps inside a record skip the range; months missing only at the start or end do not.

A range that includes zero means we can't tell yet. A trend says what changed, not why, and it is not a forecast.

Continue exploring

• City predictions: weather, climate and renewable resources
• Privacy
• Find the change. Test the evidence.
• Trend Check: is it really changing?
• Global climate atlas
• Explore the evidence
• Data sources
• Space Apps 2026: Be an Earth System Trend Detective
• Earth Pulse: where is our planet changing?
• Ground check: does NASA's record match a real thermometer?
• Cool Charlotte: where to plant trees first
• Future Cities: 20 cities toward 2050 in 3D