Ensemble dashboard

• Box – the box holds the middle half of the members (25th to 75th percentile), the line through it is the median, and the whiskers reach the 10th and 90th percentiles or the full minimum–maximum range, depending on Whisker Mode.
• Plume – the same ranges as shaded bands, with the median as a line. The easiest way to follow trends.
• Lines – each percentile as its own line, for reading off exact values.
• Spread – how wide the ensemble is over time. Rising lines mean the members disagree more and confidence is falling.
• Spaghetti – every member as a faint line. Shows whether members follow a few distinct paths rather than one broad spread.
• Scatter – one dot per member at each time. Gaps between groups of dots show the ensemble splitting into different outcomes.
• Cluster – members with similar values grouped at each time, as bubbles sized by member count.
• Scenarios – members grouped by their whole path over a chosen period.
• Patterns (beta, Premium) – GEFS, ECMWF, ECMWF AIFS and MOGREPS-G: the most likely pressure patterns over Europe as maps, with isobars and highs and lows.

Cluster groups members at each time: temperatures within ±2°C, pressure within ±4 hPa and wind gusts within ±5 mph. Precipitation and CAPE use fixed categories instead, such as Dry, Light, Moderate and Heavy. The largest group is marked as the most likely outcome. Two bubbles of similar size mean the ensemble is split between two outcomes. Hover a bubble to see every group at that time.

Scenarios compares each member’s whole path over the chosen Scenario Period, such as Days 4–7, and groups members that follow a similar course (within 2°C, 4 hPa or 5 mph on average; for precipitation and CAPE, relative to the ensemble’s own spread). Member lines are coloured by scenario, and scenario A is always the largest. Each scenario’s average line is solid inside the shaded period and dashed outside it, because the same members can spread apart again before and after. The tooltip gives each scenario’s range, and marks it as diverged once its members are more than twice as spread out as during the period.

Patterns groups whole pressure maps rather than one location. For the chosen Scenario Period, every run’s 12Z pressure maps are compared across the area shown, and runs with similar maps are grouped. The maps are first reduced to the few large-scale ways in which they differ most, and the number of patterns (up to five) is then chosen from the data: the split whose groups are most distinct is used. Each split is compared with what the same ensemble would score by chance if it had no real groups: the summary says whether the split is clear or moderate. Often the runs form one continuous spread rather than distinct outcomes; the maps then show the range of patterns the ensemble covers, and the summary says so. Runs whose maps don’t closely resemble any pattern (on average within about 3 hPa across the map, less than one isobar interval) are counted separately rather than forced into one, so at longer ranges, when the runs differ more, fewer runs count towards each pattern. Each scenario is shown as its representative member: the actual run closest to the middle of the group, so its lows and highs keep their true depth and position rather than being blurred by averaging. Isobars are drawn every 4 hPa, with H and L marking pressure centres, and the slider steps through the days of the period. Each H and L shows how many of the scenario’s runs have that centre; it is grey when fewer than half do, meaning it is a detail of a few runs rather than the consensus. The Isobars buttons switch between the representative run, the average of the scenario’s runs (smoother and shallower, showing what they share), or both together, with the average as dashed lines. The maps cover only the area stored for these models, so lows further out in the Atlantic appear only once they reach its western edge.

Member counts are estimates, not certainties: with around 30 members, groups of 15 and 13 are effectively a tie, and the tooltips say when there is no clear favourite. Scenarios are worked out separately for each chart, so scenario A for temperature is not necessarily the same members as scenario A for pressure. If up to three runs are missing data in the Scenario Period, they are left out of the grouping and drawn as amber dashed lines, and percentages are of the complete runs only.

The median splits the members into two equal halves. Where it sits within the box shows the shape of the distribution: near the bottom (more blue above), the upper half of the members is more spread out towards higher values; near the top (more green below), the opposite.

Spaghetti and Scatter also show the ensemble mean as a dashed black line. In Min/Max mode it is the mean of all members; in P10/90 mode it leaves out the most extreme 10% at each end, so a few outliers can’t drag it. Where the mean and median separate, a minority of members are forecasting something quite different.

• Show Steps – show one time of day throughout the forecast. For example, 12Z shows only midday values, making day-to-day trends easier to compare.
• Whisker Mode – switches the outer range between the 10th–90th percentiles and the full minimum–maximum.
• Quick Select – choose a city by region, or pick any grid point with the latitude and longitude dropdowns. Cities outside the chosen model’s area are greyed out.
• Download – each chart’s menu saves it as an image, including the model, run, location and variable, ready for presentations.

Charts update straight away when you change chart type, steps, height or whisker mode. On phones and tablets, zooming is switched off so that scrolling past a chart doesn’t zoom it by accident.

Each model covers the UK, Ireland and parts of Europe; the latitude and longitude lists show the exact area for the selected model.

Every location is moved to the nearest grid point, so coastal places may use a point just offshore.