Thursday, 31 August 2023
Boundary Waters (Hyatt Regency Minneapolis)
This presentation documents a recent effort to assimilate radar observations of radar Doppler velocities in a Global model at a horizontal resolution of 15 km. Radial velocities from the Canadian S-band radars are assimilated as well as Level-II data from the NEXRAD network over the continental United-States and Alaska.
Volumetric measurements from the different radar networks are first processed using the BALTRAD toolbox which computes a dedicated quality control for Doppler velocity making use of reflectivity intensity, the depolarization ratio, a median filter and exclusion of measurements over wind farms. To reduce the discrepancy between the effective resolution of the model and the observations being assimilated, the native polar measurements are averaged together and then thinned. For the averaging, an area-conserving strategy is used to arrive at “super observations” whose horizontal resolution is approximately constant. As for thinning, it is performed by giving priority to collocated observations from different radars to better constrain the two horizontal components of the winds.
Differences between the observations and their simulated counterparts (OMP’s) are examined to establish a basis for the estimation the observation errors to be used during the assimilation. The standard deviation of model-radar differences varies with range/altitude and is found to be approximately 4 m/s.
Doppler velocities are assimilated in the GPDS alongside all the other observations using a 4D-EnVar framework. The results of several forecast-only experiments are compared to find optimal values for observation errors as well as settings for the construction of super-observations and thinning. Over a two-months test period in the summer, small improvements (~0.3% reduction in RMSE for winds and temperature over North-America) could be obtained compared to a control experiment where Doppler velocities are not assimilated. No significant changes could be found over other regions of the globe.
Unfortunately, when the assimilation of Doppler velocities was attempted in a continuously cycled assimilation experiment, no significant improvements could be obtained. The reasons for this are investigated and plans for the assimilation of Doppler velocities in higher resolution predictions systems will be given.
Volumetric measurements from the different radar networks are first processed using the BALTRAD toolbox which computes a dedicated quality control for Doppler velocity making use of reflectivity intensity, the depolarization ratio, a median filter and exclusion of measurements over wind farms. To reduce the discrepancy between the effective resolution of the model and the observations being assimilated, the native polar measurements are averaged together and then thinned. For the averaging, an area-conserving strategy is used to arrive at “super observations” whose horizontal resolution is approximately constant. As for thinning, it is performed by giving priority to collocated observations from different radars to better constrain the two horizontal components of the winds.
Differences between the observations and their simulated counterparts (OMP’s) are examined to establish a basis for the estimation the observation errors to be used during the assimilation. The standard deviation of model-radar differences varies with range/altitude and is found to be approximately 4 m/s.
Doppler velocities are assimilated in the GPDS alongside all the other observations using a 4D-EnVar framework. The results of several forecast-only experiments are compared to find optimal values for observation errors as well as settings for the construction of super-observations and thinning. Over a two-months test period in the summer, small improvements (~0.3% reduction in RMSE for winds and temperature over North-America) could be obtained compared to a control experiment where Doppler velocities are not assimilated. No significant changes could be found over other regions of the globe.
Unfortunately, when the assimilation of Doppler velocities was attempted in a continuously cycled assimilation experiment, no significant improvements could be obtained. The reasons for this are investigated and plans for the assimilation of Doppler velocities in higher resolution predictions systems will be given.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner