Evidence records: showing our working

An evidence record is a machine-readable document that accompanies every data point, forecast, and alert generated by The Wind Agent. Its purpose is to provide an immutable, verifiable audit trail, detailing the exact conditions and sources that led to a particular output. This is crucial for operations where decisions are made based on wind data, allowing for post-event analysis, dispute resolution, and regulatory compliance.

Each record is structured to include metadata about the data source, processing steps, and a cryptographic signature to ensure its integrity. Key components include:

• Data Source Identification: Which numerical weather prediction (NWP) model was used (e.g., ECMWF IFS, GFS, ICON-EU).
• Model Run Details: The specific initialisation time of the model run (e.g., 2024-03-15 00:00 UTC).
• Ensemble Configuration: For ensemble forecasts, the number of members included in the calculation.
• Calibration and Post-processing: Details of any statistical calibration or bias correction applied.
• Observation Data: Identification of the synoptic stations whose measurements were incorporated for calibration or verification.
• Timestamp and Versioning: When the record was generated and the version of the processing software.
• Checksum: A cryptographic hash of the entire record, ensuring it has not been tampered with.

This comprehensive approach ensures that every piece of information presented by The Wind Agent is fully traceable and transparent.

The foundation of any forecast is the underlying numerical weather prediction (NWP) model. The evidence record explicitly states which model provided the raw data. For instance, ECMWF IFS (HRES) indicates the European Centre for Medium-Range Weather Forecasts' high-resolution deterministic model, while ECMWF ENS denotes their ensemble prediction system.

The run time specifies when the model was initialised. NWP models are typically run at 00:00, 06:00, 12:00, and 18:00 UTC. A forecast for 15:00 UTC on 2024-03-15, based on the 06:00 UTC run, means the model used atmospheric conditions observed up to 06:00 UTC to predict the future state. Older runs generally have less skill for the same forecast hour.

For ensemble forecasts, the member count is critical. A higher number of members (e.g., 51 for ECMWF ENS) provides a more robust estimate of forecast uncertainty. The record will specify 51 members for such cases.

Finally, the height at which the wind data is provided is explicitly stated (e.g., 10 m AGL, 80 m AGL). This is particularly important for the Shear Glass, where wind speeds are presented at multiple heights (10/80/120/180 m) to account for wind shear. The evidence record confirms the specific height interpolation or direct model level extraction used for each value.

For example, an entry might state: Model: ECMWF ENS, Run: 2024-03-15 00:00 UTC, Members: 51, Height: 80 m AGL. This level of detail allows users to understand the exact source and parameters of the data they are relying upon.

Raw model output often exhibits systematic biases compared to local observations. To improve accuracy, The Wind Agent applies statistical calibration to its forecasts. The evidence record details the type of calibration applied (e.g., Bias Correction, Ensemble Model Output Statistics (EMOS)).

Crucially, it also specifies the training window used for this calibration. This is the period of historical observations and corresponding model forecasts that were used to 'train' the calibration algorithm. A typical training window might be last 90 days or rolling 180 days.

For instance, if a model consistently overpredicts wind speed by 1 m/s at a specific location, the calibration algorithm learns this bias over the training window and adjusts future forecasts accordingly. The effectiveness of calibration can depend on the representativeness of the training data. For example, a calibration trained during a predominantly calm period might perform differently during a stormy spell.

Worked Example: Calibration Impact

Assume a raw model forecast for 10 m wind speed is 12.0 m/s. The evidence record indicates that a bias correction was applied, trained over the last 90 days, which found an average model overprediction of 0.8 m/s for that location and wind regime. The calibrated forecast presented to the user would therefore be:

12.0 m/s (raw) - 0.8 m/s (bias) = 11.2 m/s (calibrated)

This adjustment, and the parameters used to derive it, are fully documented in the evidence record.

For calibration and for providing real-time observations, The Wind Agent integrates data from a network of synoptic weather stations. The evidence record lists the specific stations whose data contributed to the particular output. This includes both stations used for calibrating forecasts and those providing the 'measured' component of an observation versus model comparison.

In Ireland, these typically include stations operated by Met Éireann, such as Dublin Airport (EIDW), Roches Point (EICK), or Mace Head (EIIM). For Northern Ireland, stations from the UK Met Office might be listed, such as Aldergrove (EGAA).

Listing these stations provides transparency regarding the observational basis of the data. It allows users to cross-reference with publicly available data from these stations if desired. It also highlights the spatial representativeness: if a forecast for a specific location is calibrated using a station 50 km away, this context is provided.

For example, an evidence record might state: Observed data from: Dublin Airport (EIDW), Mace Head (EIIM). This indicates that the calibration algorithm incorporated historical data from these two stations, or that a recent observation presented was drawn directly or indirectly from them. The Agreement Spine, for instance, explicitly compares model output to nearby observations, and the evidence record would list the contributing stations.

Every evidence record includes a precise timestamp indicating when it was generated. This ensures that the record itself is auditable and can be linked to specific events or decisions. The timestamp is typically in UTC (Coordinated Universal Time) to avoid ambiguity with local time zones.

The version of The Wind Agent's processing software is also recorded. As algorithms and data pipelines evolve, versioning ensures that any changes to the methodology are documented. If an issue is identified, knowing the software version allows for precise identification of the code that produced a particular output.

Perhaps the most critical component for data integrity is the checksum. This is a cryptographic hash (e.g., SHA-256) of the entire evidence record. A checksum acts as a digital fingerprint. If even a single character in the record is altered, the computed checksum will change, immediately indicating tampering. This provides a robust mechanism to verify the authenticity and integrity of the evidence record.

Worked Example: Checksum Verification

Imagine an evidence record record.json is generated, and its SHA-256 checksum is e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855. If this record is later retrieved and its checksum is re-calculated, and the new checksum matches the original, you can be certain the record has not been modified. If it differs, the record's integrity is compromised.

This combination of timestamp, version, and checksum guarantees the reliability and trustworthiness of the data provided by The Wind Agent.

Before making a critical operational decision based on The Wind Agent's data, reviewing the associated evidence record is a best practice. While the instrument presents clear, actionable information, the evidence record provides the underlying context and confidence metrics.

1. Check the Model and Run Time: For time-sensitive operations, ensure the forecast is based on the latest available model run. An older run might not capture recent atmospheric changes.
2. Review Calibration Details: Understand if and how the data has been calibrated. If the training window was short or unrepresentative, this might subtly influence confidence in extreme events.
3. Note Observation Stations: If using data for a specific microclimate, consider the proximity and representativeness of the listed observation stations. A station far away or in a very different terrain might not perfectly reflect local conditions.
4. Examine Ensemble Spread (if applicable): For ensemble forecasts (e.g., the Exceedance Fan), the evidence record implicitly supports the spread by detailing the number of members. A wider spread, even with calibration, indicates higher uncertainty. The Exceedance Fan (chart fan) directly visualises this by showing probabilities of exceeding your limit across the ensemble.
5. Verify the Checksum: For critical audits or legal purposes, always re-calculate and compare the checksum to ensure the record's integrity.

By systematically reviewing these elements, users can gain a deeper understanding of the forecast's robustness and make more informed decisions, aligning with their own operational procedures and risk assessments.

In many industries, wind conditions are critical parameters for safety, operational efficiency, and contractual obligations. When incidents occur, or when performance is questioned, having an indisputable record of the environmental conditions is paramount. This is where The Wind Agent's evidence records become invaluable.

• Dispute Resolution: If a project experiences delays or damage attributed to wind, the evidence record provides a neutral, verifiable account of the wind conditions at the time. This can be crucial in contractual disputes with insurers, clients, or contractors. For example, if a crane operation was halted due to high winds, the record proves the wind speed and gust at the working height, as per the forecast or observation, and its provenance.
• Regulatory Compliance: Many regulatory bodies (e.g., aviation authorities, health and safety executives) require robust documentation of environmental conditions for certain operations. The detailed, verifiable nature of the evidence records assists organisations in demonstrating compliance.
• Internal Audits and Post-Mortems: For internal safety reviews or performance analyses, evidence records allow teams to accurately reconstruct past conditions, identify contributing factors, and refine operational procedures.
• Insurance Claims: For insurance purposes, clear and verifiable data on wind conditions can significantly expedite claims processing and help establish liability.

The cryptographic checksum ensures that these records are tamper-proof, providing a level of trust that simple screenshots or manual logs cannot match. This 'showing our working' approach builds confidence and reduces operational risk.

The Wind Agent maintains a comprehensive archive of all generated evidence records. This long-term retention policy ensures that historical data, along with its full provenance, remains accessible for future reference, audits, and analysis.

Retention periods are designed to meet typical industry and regulatory requirements, often extending for several years. This allows users to retrieve records for events that occurred months or even years in the past, without needing to store them locally. The storage infrastructure is designed for resilience and data integrity, ensuring that records are protected against loss or corruption.

Users can typically access these historical records through their account interface, linking directly to past forecasts, alerts, and observations. This capability is particularly useful for seasonal planning, long-term trend analysis, or revisiting specific operational periods.

For example, if a wind farm needs to analyse turbine performance during a specific storm event from two years prior, they can retrieve the exact wind forecasts and observations, complete with their evidence records, to understand the conditions that prevailed. This ensures continuity and accountability over extended operational lifecycles.

This commitment to long-term data retention, combined with the integrity features of the evidence record, underpins the reliability and trustworthiness of The Wind Agent as a critical operational instrument.