Written by Gerard Castro, CTO at Nebbo.
Understanding seasonal forecasting requires familiarity with specialized terminology. This glossary provides clear definitions of key technical terms used in seasonal weather forecasting, helping users better interpret and utilize forecast information provided by Nebbo and other meteorological services.
Fundamental Concepts
Baseline Period
The historical time period used as a reference for calculating averages, anomalies, and other statistical measures. This period typically spans multiple decades to capture climate variability and establish what is considered “normal” for a particular location and time of year.
In particular, the WMO defines as period for the climatological standard normals as any consecutive period of 30 years starting on 1 January of a year ending with the digit 1: e.g 01-01-1991 to 31-12-2020.
Climatology
The long-term average of meteorological variables (such as wind speed or temperature) calculated over a specific baseline period. Climatology serves two primary purposes: as a simple predictive tool based on historical patterns and as a reference for calculating anomalies. While useful for understanding typical conditions, climatology alone cannot account for trends, cycles, or irregular variations in weather patterns.
Anomaly
The deviation of a meteorological variable from its climatological average for a specific time and location. Anomalies are typically expressed as differences (e.g., +2°C) or percentages (e.g., 15% above normal) relative to the climatological value. Positive anomalies indicate values above the average, while negative anomalies indicate values below the average.
Absolute
The actual measured or predicted value of a meteorological variable, as opposed to an anomaly. For example, an absolute wind speed of 5 m/s represents the actual wind speed, while an anomaly of +1 m/s indicates the wind speed is 1 m/s above the climatological average.
Forecast Terminology
Initialization Date
The specific date when a forecast model begins its simulation. This date marks when the model incorporates the most recent observational data as its starting point. The initialization date is crucial for understanding how current the data feeding into the forecast is.
Since seasonal models are composed by ensembles of predictions (which may have different initialization dates), a “nominal” start date is often used as initialization date.
Forecast Date
The future date or period for which a prediction is made. In seasonal forecasting, forecast dates typically refer to entire months or seasons rather than specific days.
Lead Month
Lead time is, in general, the interval between the initialization date and the forecast date. Nevertheless, following C3S & ECMWF convention, lead month is the number of complete calendar months between initialization and forecast dates.
For instance, a forecast issued at the beginning of a month and which is valid at the end of the month is considered lead month 1; then, if it was valid by the end of next month, that prediction would have lead month 2 and so on.
Prediction Centre
A meteorological organization that develops and operates forecast models. Major prediction centers include the European Centre for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP), UK Met Office (UKMO), and others. Each center maintains its own forecast models with unique characteristics and methodologies.
Observational reference
The meteorological conditions considered as “observed” or occurred that serves as the reference against which forecast performance is measured. Often equally referred as “ground truth” too.
This data comes from reliable observational sources such as weather stations, satellites, reanalysis datasets or synthetic data as Vortex SERIES.
Confidence
The degree of certainty in a forecast, often expressed as a probability or confidence level. Higher confidence indicates greater certainty that the forecasted conditions will occur. Confidence typically decreases with longer lead times due to the inherent uncertainty in long-range predictions.
Ensemble Forecasting
Ensemble
A collection of multiple forecast simulations (called members) for the same time period, created by running a model with slightly different initial conditions or model parameters. Ensembles help quantify uncertainty by showing the range of possible outcomes rather than a single deterministic prediction.
Ensmean
Short for “ensemble mean,” this is the average of all ensemble members for a particular forecast. The ensemble mean often provides a more reliable forecast than any individual ensemble member, as it smooths out random variations while preserving the signal common to most members.
Lower/Upper Terciles
Divisions that split the range of possible values into three equal parts based on historical data. The lower tercile contains the lowest third of historical values, the middle tercile contains the middle third, and the upper tercile contains the highest third. Seasonal forecasts often express probabilities of conditions falling within each tercile (e.g., “40% chance of above-normal temperatures” means a 40% probability of conditions in the upper tercile).
Uncertainty
The range of possible outcomes in a forecast, typically represented by the spread of ensemble members or through statistical measures like prediction intervals. Greater uncertainty indicates a wider range of possible outcomes and typically lower confidence in any specific outcome.
Interval Coverage and Width
Interval coverage is the percentage of times that observed values fall within a prediction interval. Ideally, a 95% prediction interval should contain the observed value 95% of the time. Interval width is the size of the prediction interval, with narrower intervals indicating more precise forecasts. There is typically a trade-off between coverage and width—wider intervals provide better coverage but less precision.
Validation and Calibration
Validation
The process of evaluating forecast performance by comparing predictions to observed outcomes (ground truth). Validation metrics may include measures of accuracy, bias, reliability, and skill relative to reference forecasts like climatology.
Model calibration
Model calibration or model postprocessing is the statistical adjustment of raw model’s outputs to correct for systematic biases and improve reliability using the observational reference. Calibration typically involves comparing historical forecasts with observations to develop correction factors that are then applied to new forecasts.
Observational Reference Calibration
Observational reference calibration specifically adjusts reanalysis or synthetic observational data by comparing it to observed outcomes, directly addressing biases relative to real-world measurements.
Advanced Concepts
Multimodel
An approach that combines forecasts from multiple independent prediction models to create a more robust forecast. Multimodel forecasts often outperform individual models by leveraging the strengths of each model and reducing the impact of model-specific biases.
Reanalysis
A dataset created by combining historical observations with a consistent modern forecast model to produce a comprehensive, physically coherent record of past weather conditions. Reanalysis datasets provide a gridded, gap-free representation of historical conditions and are often used as reference data for model development and validation.
Teleconnection
A statistical correlation between weather phenomena occurring in widely separated regions of the globe. Teleconnections, such as the El Niño-Southern Oscillation (ENSO) or the North Atlantic Oscillation (NAO), can provide predictability for seasonal forecasts by linking distant weather patterns through atmospheric or oceanic circulation.
Hindcast Period
A period in the past for which a forecast model is run to evaluate its performance against known outcomes. Hindcasts (also called retrospective forecasts) are crucial for understanding model biases, calibrating forecasts, and assessing forecast skill under various conditions. In particular, C3S sets 1993-01 to 2016-12 as fixed hindcast period.
Operative Period
The current period during which a forecast model is being used to generate real-time predictions for future conditions. The operative period follows the hindcast period (but is not necessarily consecutive) and represents the actual application of the model for forecasting purposes.
Downscaling
The process of translating coarse-resolution global model forecasts to finer spatial scales to provide more localized predictions. Downscaling can be statistical (using empirical relationships between large-scale and local conditions) or dynamical (using higher-resolution regional models nested within global models).
Summary
This glossary covers essential technical terms used in seasonal forecasting, from fundamental concepts like baseline periods and anomalies to advanced topics like multimodel approaches and downscaling. Understanding these terms helps users interpret forecast information more effectively, recognize the capabilities and limitations of seasonal predictions, and make more informed decisions based on forecast data. As seasonal forecasting continues to evolve, familiarity with this terminology provides a foundation for engaging with both current and emerging forecast products.
