NWP and Reality
Owing to the non-linear nature of the forecast system, output from NWP models does not behave in a simple or regular way. Errors can be due to:
- The inherent structure of the NWP model.
- Errors in the initial handling of information and data.
- Propagation and non-linear growth of analysis errors moving downstream.
NWP model forecasts are true to the physical representation and parameterisation of the atmosphere and its processes within the NWP model programs and systems but minor inconsistencies in initial data can and do amplify. This is the whole point of the ensemble forecast so that uncertainty may be investigated. But users should be aware of how best to deal with error growth and to understand how forecaster modification can improve forecasts.
Model forecasts rely heavily on the data available, particularly the addition of new and up-to-date data that can be added to the model background fields. Users should be alert to significant differences between the observed data and background fields (analysis increments) as this points out one particular area where the model expectations may be incorrect - there may be others that are not identified in this way. Users then have to decide whether the data or the background is anomalous and incorrect and then later assess whether the model has adapted, at least partially, to the latest data. Adjustment to incorrect data, or non-fitting of correct data, both have the potential to amplify/cause forecast errors.
Finally, meteorological disturbances and the associated energy propagate downstream (downstream spread of influence) both within NWP models and in reality. Anomalies in the analysis will therefore affect forecasts well downstream during the forecast period - some differences weakening away, but others amplifying to yield a very incorrect result. Users need to be alert to the development and advance of disturbances that may potentially amplify and that arise from observed data departing from background fields.
Overview
Community Forums
Content Tools