Tercile and quintile probabilities explained

How below-normal, near-normal and above-normal probabilities are derived from ensemble members and hindcasts, and how to read the tercile summary map.

Written and maintained by seasonforecasts · Published 11 September 2026 · Reviewed 12 September 2026 · Sources: C3S seasonal forecast catalogue and product descriptions

Categories from the hindcast

A tercile forecast divides the model’s possible outcomes into three nominal thirds: below normal, near normal and above normal. Threshold ties can make their actual hindcast frequencies unequal. The boundaries are set from the hindcast: for each start month, lead and grid cell, all qualifying hindcast members from 1993 to 2016 are pooled and the 33⅓ and 66⅔ percentiles are taken as thresholds. Five quintile categories are separated by the 20th, 40th, 60th and 80th percentiles. This site shows only the lowest and highest quintiles, using the 20th and 80th percentiles; the median exceedance uses the 50th.

Counting members

The probability of a category is the share of forecast members that land in it. If 30 of 50 members are above the upper threshold, the upper-tercile probability is 60%. Here the lower tercile uses a strict “less than” and the upper tercile a strict “greater than”; equality with either tercile threshold belongs to the middle tercile. The lowest/highest quintile and above-median events also use strict comparisons. The three tercile probabilities sum to 100% before rounding; stored or displayed values can differ slightly from that total.

Reading the tercile summary

The tercile summary map condenses the three probabilities into one colour. A cell is coloured for above or below normal only when that outer tercile exceeds 40% and is strictly more probable than both other categories. Any tie for the highest probability stays neutral, including 10% below / 45% near / 45% above. A middle-dominant distribution such as 12% below / 45% near / 43% above also stays neutral. Neutral means “No favoured outer tercile”; it does not by itself predict near-normal conditions and is distinct from missing data. Exact 40% is neutral; 50%, 60% and 70% enter the next stronger band on either side. Inspect the individual tercile probabilities when the distinction matters.

This follows the ECMWF summary definition (page 4), with an explicit local policy of leaving winning ties neutral.

The baseline is not zero

The nominal climate references are 33.3% per tercile, 20% per outer quintile and 50% above the median. These are not guaranteed local hindcast frequencies: finite samples and tied values matter. Ties at the tercile boundaries go to the middle category; ties at quintile or median thresholds are excluded from their strict outer-tail and above-median events; members are never randomly split or jittered. For an all-zero precipitation reference, achieved lower/near/upper frequencies are 0/100/0%, and strict above-median frequency is 0%. For an illustrative reference of seven zeros and three ones, tercile frequencies are 0/70/30%, with 30% above the median. These synthetic examples explain threshold ties; they are not operational forecasts. A 60% category forecast describes the period mean, not rain on 60% of days or a guarantee that such forecasts verify 60% of the time.

How this site derives them

Probabilities here are computed locally from individual forecast and hindcast members downloaded from C3S, using the C3S approach of counting members against hindcast categories. This implementation uses linearly interpolated sample quantiles and the strict threshold rules above. Member counts follow C3S guidance on qualifying members for each centre. The local quantile, tie, masking and display choices are not a claim of exact equivalence to Copernicus charts. A derived cell requires all selected forecast and hindcast members in every constituent month. See the C3S product descriptions for the reference definitions, and why a probability is not forecast skill before treating any percentage as confidence.

Sources and method

These references explain the underlying science and provider conventions. Descriptions of this site’s calculations, colours and coverage were checked against its processing code and imported data.