Methods, data and limits

How to read a trend

What the numbers mean

A trend is the long-term direction in a monthly or seasonal record. Sen’s slope reports the typical rate of change per decade. Mann–Kendall tests whether the values tend to move consistently up or down. A confidence interval gives a plausible range for the slope.

Several checks

We compare month-by-month, anomaly, calendar-month and annual-season methods. Autocorrelation can make nearby months less independent; Hamed–Rao adjusts the test’s uncertainty. A window sweep checks whether the conclusion depends on the chosen years.

Data source

NASA POWER provides gridded reanalysis and satellite-derived products. Regional values are means of grid cells. These are model-assisted estimates, not direct measurements at every location.

Limits

A significant trend does not show what caused it. A weak result does not prove no change occurred. Window comparisons overlap and are not independent tests. Data availability and observing systems change through time.

Validation

Statistics are cross-checked against Python pymannkendall and synthetic series with known answers. Each result includes diagnostics so you can assess gaps, unusual years and autocorrelation.

Credits

Built with AI coding tools — see repository DISCLOSURE. NASA POWER data access and statistical methods are credited in each result export.