MerQuR+Statistics Glossary

Statistics Glossary → Seasonal Kendall test

Seasonal Kendall testconcept

The seasonal extension of the Mann-Kendall test, which tests for a monotonic trend in seasonal time series by comparing each season only with itself.

The Mann-Kendall test looks for an increasing or decreasing trend by counting, over all pairs of observations, how often the later value is larger or smaller than the earlier one, and it needs no distributional assumption. If the series is strongly seasonal (summer months always high, winter months always low), these comparisons get mixed up with seasonal differences. The Seasonal Kendall test computes the Mann-Kendall statistic separately for each season (for example each month), comparing Januaries only with Januaries and Julys only with Julys, and sums these statistics into an overall trend test.

The size of the trend is summarized by the seasonal Sen slope, the median of all year-to-year slopes within the same season. The test assumes the trend runs in the same direction in every season; if some months rise and others fall, they can cancel out, so the homogeneity of trends across seasons should be checked separately. Its basic form assumes observations are independent across seasons; with strong serial dependence, versions with a corrected variance are needed.

An everyday example

Water temperature in a lake is measured every month for 15 years. Because summer and winter differ by as much as 15 °C, the slight warming from year to year is buried in the seasonal swing in a plain Mann-Kendall test. The Seasonal Kendall test compares each month with its own history and finds a common warming trend across all months; the seasonal Sen slope puts it at about 0.05 °C per year.

In MerQur

In Mann-Kendall + Sen's Slope, the 'Seasonal Kendall period (0 = off)' option compares each season within itself and computes the seasonal Sen slope.

Analyses where this concept appears

Mann-Kendall + Sen's Slope

Related concepts

Sen's slopeSeasonality and seasonal period