MMM SEMINAR NCAR
Object-based evaluation of weather
forecasts: Application to NWP models
UCAR/MMM
One of the great challenges created by ever-increasing resolution of numerical forecasts is the difficulty in quantifying their realism and accuracy. Real events are often highly localized and episodic and forecasts are therefore susceptible to errors in timing or placement that can lead to large errors in traditional metrics of skill. Such errors are often particularly large for models that explicitly represent localized weather events.
Object-based methods of verification, described in a related seminar by Barbara Brown (Aug. 23), are applied to two problems: local wind circulations forced by complex terrain and warm season rainfall systems. Objects are defined in varying complexity in space-time, ranging from one-dimensional objects in time series to mesoscale rain systems in three dimensions (x,y,t). The resolution-dependence of forecast skill in complex terrain is assessed by defining objects as relatively rapid wind changes in time series and investigating their correspondence in MM5 forecasts and observations. From this analysis, it is demonstrated that finer resolution is capable of resolving diurnal and terrain induced motions statistically, but only marginally in a deterministic sense. From the analysis of warm season rain systems, biases of position, timing, intensity and size are descerned using 4-km WRF forecasts. While many attributes of rain systems are well represented, the model tends to produce systems that last too long and are too large.
Finally, extensions of object-based techniques to ensemble forecasts and climate simulations are discussed.
.
Refreshments
NCAR-Foothills Laboratory
Bldg 2 Auditorium (Rm1022)