Numerical Forecasting Model Errors

               ---------------------------------------------

Numerical model forecast ability is limited by errors originating in the
model initialization, model imperfections, and the inherent growth of
errors. Errors common to numerical weather prediction products are usually
associated with data and analysis errors, and specific model weaknesses:

Data and analysis errors arise from the lack of conventional observations
or satellite derived data, and errors found in the observations themselves.
These produce an imperfect analysis for model initialization which enhances
errors in the early forecast period. Defects in the analysis scheme, such
as a poor first-guess input, reduce the accuracy of the initialization
process by decreasing the effect of correct data.

Model weakness errors result from a specific model's overall skill and
characteristics. These tend to reduce forecast accuracy in a systematic
manner and lose skill with increasing forecast time. Model limitations may
include the model's horizontal and vertical resolutions, physical
parameterizations, model initialization and numerical scheme.

Model errors and trends found in FNMOC numerical analysis and forecast
products, as well as those of other numerical weather prediction centers,
are commonly referred to as:

Systematic model errors refer to forecast errors which occur routinely in
the same location and/or time. For example, 1000 mb forecast heights are
typically too low over the Philippine Sea. Systematic errors usually grow
in amplitude through the forecast period. The similarity to errors in the
model climatology over extended periods indicates these errors represent a
gradual drift from the climate of the atmosphere to that of the model. The
rate of model drift may vary from season to season, but the overall error
appears to be independent of the initial data.

Model tendencies refer to the similar handling of a synoptic forecast
situation without regard to location or season, and are usually related to
data density. For example, "Developing mid-latitude maritime lows tend to
be slightly underforecast and slow to deepen. Filling, mature lows are
overforecast and slow to fill. Deepening and filling lows show no
consistent directional or positional bias."