METEOPROOF - Model Verification & Validation

📊 Daily Model Verification

Compare model performance and forecast evolution per run timestamp against observations.

Quick Date Navigation

💡 How to read this chart?

Show more ▼

This chart shows how the forecast for a specific target date changes as the calculation run approaches (from left to right, from D-13 to D-0).

Each step on the X-axis represents a specific run timestamp. (Example: D-9_12z is the forecast from the 12:00 UTC run generated 9 days prior to this target date).

The cyan dotted line (Cluster Consensus) calculates the median of all models per run to filter out extreme outliers, running through to the end of the period to show the stable trend.

Multimodel Accuracy & Bias Verification

Target Date: -- | Location: -- (Max)

Reference: --. Calculated RMSE and Bias values (AVG D1-6 = average for D-1 through D-6):

RMSE Scale (°C):

METEOPROOF.COM

💡 What is Accuracy & Bias Verification?

Show more ▼

This figure displays the Root Mean Square Error (RMSE) in degrees Celsius, relative to the reference (Official observation). The lower the number (green), the more accurate the model was.

Bias D1-6: 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 D-1 through D-6: ▲ red means the model runs structurally too warm, ▼ blue means too cold, ● means no clear direction. A model can have a solid RMSE and still carry a real bias -- the two rows answer different questions.

Run-to-Run Stability

Target Date: -- | Location: -- (--)

How much a model's forecast changes between consecutive runs on each lead day (lower = more consistent).

RMSE Scale (°C):

METEOPROOF.COM

💡 What is Run-to-Run Stability?

Show more ▼

Instead of accuracy, this table measures consistency: how much a model's forecast for the same target date changes between one run and the next, on a given lead day. A model can be accurate on average but still swing wildly from run to run -- this table catches that.

Example: if a model forecasts 20°C, then 21°C, then 19°C across three consecutive runs, the run-to-run changes are +1°C and -2°C. The score is the root-mean-square of those changes: √((1² + 2²) / 2) ≈ 1.58°C.

Only runs that actually published a value are compared (missing/skipped runs are ignored, never counted as a "jump"). The lower the number (green), the more consistent the model was from run to run. Columns are ranked best to worst, left to right, based on AVG D1-6 (the average over D-1 through D-6, shown in the top row).