The animations below show how a simple 2D shallow water model can be used to test the impact of different algorithms used in numerical weather prediction. Experiments 4, 5, and 6 show the impact of a single "bad" observation at the initial time. Each simulation is initialized using two balanced vorticies of differing magnitudes that are spaced 15 gridpoints apart. The grid spacing is 40 km. The southern boundary is located at 15 N latitude. A tracer is included in the initial conditions, in this case, mixing ratio which decreases in magnitude as you go north. In a perfect simulation the vorticies would hold their magnitudes and drift northwest due to the coriolis affect.
Each simulation generates a 72hr forecast. Comparisons that illustrate the impact of using various model configurations and of errors in the initial fields are presented below.
72-hr simulation, no horizontal smoother, no time filter. Fields become noisey.
72-hr simulation, 2nd order horizontal diffusion, time filter. Horizontal diffusion erodes the initial vortex throughout the forecast.
72-hr simulation, 6th order horizontal filter, time filter. The filter preserves the vortex.
72-hr simulation, 6th order horizontal filter, time filter, bad height observation. The model recovers quickly.
72-hr simulation, 6th order horizontal filter, time filter, bad wind observation. Gravity waves are generated as the model reconstructs the mass-wind balance in the vicinity of the bad wind observation.
72-hr simulation, no horizontal smoother, time filter, bad wind observation.
72-hour forecast, 2D horizontal diffusion versus 6th order filter
3-hour forecast, bad height observation versus control with 6th order filter
3-hour forecast, bad wind observation versus control with 6th order filter
72-hour forecast, bad wind observation: No filter versus control with 6th order filter
Please refer questions or comments to the CRASmaster: cras@ssec.wisc.edu