The Agreement Spine: how models agree or disagree on the forecast

The Agreement Spine is a horizontal strip chart, segmented by hour or by three-hour blocks, that uses colour to indicate the level of consensus among the underlying forecast models. It provides an immediate visual cue for periods when the forecast is stable and consistent, versus times when models diverge significantly.

Each segment of the spine represents a future time point. The colour of the segment indicates the degree of agreement across the various models and ensemble members contributing to the forecast. This allows you to quickly scan a multi-day forecast and identify hours or days where the models are in strong alignment, suggesting a higher confidence in the forecast, or where they are in disagreement, indicating greater uncertainty.

This visual summary is particularly useful for operational planning, as it highlights times when decisions might need to be deferred, or when a more conservative approach to limits is warranted. It acts as a concise summary of the more detailed model comparison and ensemble plume charts, distilling complex information into an easily interpretable form.

The Agreement Spine typically uses a colour scale to represent the level of consensus:

• Green: Indicates strong agreement. The primary models and a high fraction of ensemble members are forecasting similar wind speeds and directions. This suggests a high confidence in the forecast.
• Yellow/Amber: Represents moderate agreement. There might be some spread among the models or ensemble members, but a general trend or range is still discernible. Caution may be advised, and deeper investigation might be beneficial.
• Red: Signifies significant disagreement or a 'split' in the forecast. Models are predicting substantially different outcomes for wind speed or direction. This is a strong indicator of high uncertainty, often associated with complex meteorological situations such as developing low-pressure systems, frontal passages, or convective activity.

The thresholds for these colour states are internal to The Wind Agent and are based on statistical measures of spread, such as standard deviation or inter-quartile range, across the model outputs. For example, a 'green' state might correspond to all primary models being within 10% of each other for speed and within 20 degrees for direction, while 'red' might indicate differences exceeding 30% for speed or 60 degrees for direction. Your own operational risk appetite will govern how you interpret these states.

The Wind Agent calculates agreement by comparing the forecasts from multiple global and regional models, such as ECMWF, GFS, ICON, and UKMO, along with their respective ensemble members where available. For each forecast hour, the system evaluates two primary metrics:

1. Wind Speed Consensus: This assesses how close the predicted wind speeds are from each model and ensemble member. A common method is to calculate the range or standard deviation of speeds across all sources. For instance, if four models predict 10, 11, 12, and 10 m/s, the range is 2 m/s. If another set predicts 5, 10, 15, and 20 m/s, the range is 15 m/s, indicating much lower agreement.
2. Wind Direction Consensus: This evaluates the spread in predicted wind directions. Due to the circular nature of direction (e.g., 350° and 10° are close, not far apart), specialised circular statistics are used. The system might calculate the mean resultant vector or circular standard deviation. For example, if models predict 270°, 280°, 260°, and 275°, the agreement is high. If they predict 90°, 180°, 270°, and 0°, agreement is very low.

These two consensus scores are then combined, often with weighting, to produce a single agreement score for that time step, which is mapped to the green, yellow, or red colour. The weighting might prioritise speed agreement for some applications and direction for others. For example, a difference of 5 m/s in a 10 m/s forecast is more significant than a 5 m/s difference in a 30 m/s forecast, and the scoring accounts for this relative difference.

It is crucial to understand that agreement does not equate to accuracy. A forecast where all models agree strongly (green spine) is not necessarily accurate. It simply means that the models are converging on a similar solution. This can happen even when all models are wrong, for example, if they all share a common bias or miss a small-scale feature.

Conversely, a forecast with significant disagreement (red spine) does not automatically mean the forecast is incorrect. It indicates high uncertainty, suggesting that the atmosphere's state is complex and difficult for models to resolve consistently. In such cases, one of the diverging forecasts might turn out to be accurate, but it is impossible to know which one in advance.

The Agreement Spine is a measure of forecast confidence, not forecast skill. High agreement suggests higher confidence in the forecast range, while low agreement suggests lower confidence. It is a tool to help you assess the reliability of the forecast, rather than its absolute correctness. Always cross-reference with observations when available to understand model performance in your specific location.

Disagreement among models can manifest in different ways, and the Agreement Spine helps to distinguish between them:

• Disagreement on timing: Models might agree on the overall wind speed or direction but differ on when a change will occur. For example, one model might predict a front passing at 14:00, while another predicts it at 17:00. This often appears as a 'red' segment on the spine that shifts along the timeline when comparing different model runs.
• Disagreement on level: Models might agree on the timing of an event but differ significantly on the magnitude of the wind speed or the exact direction. For instance, all models might show a peak wind at 15:00, but one predicts 15 m/s, another 25 m/s. This will also result in a 'red' segment, but the peak will be aligned across models.

Consider an example: for a specific hour, Model A forecasts 10 m/s from 270°, Model B forecasts 12 m/s from 280°, and Model C forecasts 20 m/s from 320°. The speed range is 10 m/s (from 10 to 20 m/s), and the direction range is 50° (from 270° to 320°). This significant spread would likely trigger a 'red' state on the Agreement Spine, indicating high uncertainty in both speed and direction. Understanding the nature of the disagreement (timing vs. level) often requires examining the individual model traces on the model_compare chart.

The Agreement Spine is a valuable tool for operational decision-making, particularly when dealing with wind-sensitive activities. When the spine shows 'red' for a future period, it signals a heightened risk due to forecast uncertainty. In such situations, consider the following:

• Delaying decisions: If possible, postpone critical decisions until the forecast models converge. Uncertainty often decreases as the forecast lead time shortens and more observations become available to the models.
• Applying stricter limits: If operations must proceed, consider applying more conservative wind speed or direction limits. For instance, if your normal limit is 15 m/s, you might reduce it to 10 m/s during periods of high uncertainty.
• Increased monitoring: Plan for more frequent checks of the forecast and live observations as the uncertain period approaches. The Wind Agent's alerts can be configured to notify you of significant changes.
• Contingency planning: Prepare for a wider range of potential wind conditions. This might involve having alternative operational plans or equipment ready.

For example, if The Wind Agent shows a 'red' Agreement Spine for a period when a crane lift is scheduled, it indicates that there is substantial disagreement among models regarding the wind conditions. This is a clear signal to re-evaluate the lift schedule, potentially deferring it or implementing additional safety margins, rather than proceeding based on a single model's output.

The Agreement Spine and the exceedance fan are complementary features within The Wind Agent, providing different but equally important perspectives on forecast uncertainty.

• The Agreement Spine focuses on the consensus among models and ensemble members. It tells you how much the forecasts differ from each other.
• The exceedance fan (P10–P90 bands) shows the probability of exceeding your specified limit at your working height. It tells you what fraction of ensemble members predict conditions above your threshold.

When the Agreement Spine is 'green', it suggests that the ensemble members are tightly clustered, and the exceedance fan will likely be narrow. This indicates high confidence in the forecast and a clearer picture of the probability of exceeding your limit. If the fan shows a low probability of exceedance during a 'green' period, you can proceed with higher confidence.

When the Agreement Spine is 'red', the ensemble members are widely spread. The exceedance fan will likely be broad, indicating a wide range of possible outcomes and thus higher uncertainty in the probability of exceeding your limit. For example, a broad fan might show a P10 (10% chance) of exceeding 10 m/s, but a P90 (90% chance) of exceeding 5 m/s, with the models disagreeing significantly on the actual speed. In such a scenario, even if the median forecast is below your limit, the wide spread on the fan, coupled with the red spine, indicates a high-risk situation requiring careful consideration.