METEOPROOF

Calendar Days & Coverage: The table includes every calendar day of the selected month. By default only the Average and Best/Worst Performance rows are shown -- click "Show all days" above the table to expand the full day-by-day breakdown and see exactly when a model had a strong or weak day. Click any date to instantly jump to its Daily Verification view.

Unverified Data Warning: Days marked with an amber asterisk (*) indicate that the official observation is not yet available, and the model median is used as a temporary reference.

Forecast Horizon Selection & Filtering:
• D1-6: Average calculated over D-1 t/m D-6 (including all models, handling missing terms dynamically).
• D7-13: Average calculated over D-7 t/m D-13 (excluding ICON due to limited data).
• TOTAL: Overall average across all days D-1 t/m D-13 (excluding ICON and UKMO to prevent bias from limited horizons).

Color Coding & Performance: Each cell displays the Root Mean Square Error (RMSE) in degrees Celsius for a specific model on that day, color-coded according to accuracy.

Bias (Month): a separate row with its own blue/red color scale, on purpose -- RMSE only tells you how big the error is, never which direction. Bias is the plain average of (forecast − observed) over the whole month: ▲ red means the model runs structurally too warm, ▼ blue means too cold, ● means no clear direction.

Forecast Horizons (D-1 to D-13): Each row aggregates one specific lead time across the whole month, so you can see how model skill degrades (or holds up) the further in advance a forecast was made.

What does "D-6" mean? It's the forecast issued 6 days before the target date -- both its 00z and 12z runs. The D-6 row pools every 00z/12z forecast made at that lead time, on every day of the month, into a single RMSE per model.

Worked example: Say AIFS's D-6 forecasts were checked against the actual max temperature on two different days:
• Target 05/08 (actual 24.0°C): 00z run → 24.8°C (error +0.8°C), 12z run → 23.6°C (error -0.4°C)
• Target 06/08 (actual 21.0°C): 00z run → 22.0°C (error +1.0°C), 12z run → 20.5°C (error -0.5°C)
All 4 errors are squared and pooled together: (0.8² + 0.4² + 1.0² + 0.5²) / 4 = 0.51, then RMSE = √0.51 ≈ 0.72°C -- that's the number that would show up in AIFS's D-6 cell.

Model Coverage at Longer Leads: Not every model publishes forecasts all the way out to D-13 -- ICON and UKMO in particular stop appearing around D-6/D-7 in this dataset. Where a model has no data for a given lead, its cell simply shows "None" and it's skipped when picking that row's Best/Worst Model.

D1-6 Bias / TOTAL Bias: a separate row under each AVG row, with its own blue/red color scale on purpose -- RMSE only tells you how big the error is, never which direction. Bias is the plain average of (forecast − observed) over that same window: ▲ red means the model runs structurally too warm, ▼ blue means too cold, ● means no clear direction. TOTAL Bias excludes ICON and UKMO for the same coverage reason as TOTAL AVG.