Hybrid En3DVar Radar Data Assimilation and Comparisons with 3DVar and EnKF with OSSEs and a Real Case

Hybrid En3DVar Radar Data Assimilation and Comparisons with 3DVar and EnKF with OSSEs and a Real Case


Kong, Rong
Ph.D. Student


November 17, 2017 - 2:00 pm


November 17, 2017 - 3:00 pm


120 David L. Boren Blvd., Room 5600, Norman, OK 73072   View map

Studies have shown benefits of hybrid ensemble-variational data assimilation (DA) over pure ensemble or variational algorithms. Such advantages, if any, for convective-scale radar DA have not been clearly demonstrated, however. A hybrid ensemble-3DVar (En3DVar) system is developed recently based on the ARPS 3DVar and EnKF systems at CAPS. Hybrid En3DVar is compared with 3DVar, EnKF, and pure En3DVar for radar DA through observing system simulation experiments (OSSEs) under perfect and imperfect model assumptions. It is also applied to a real case including multiple tornadic supercells. For the real case, radar radial velocity and reflectivity data are assimilated every 5 minutes for 1 hour that is followed by short-term forecasts. DfEnKF that updates a single deterministic background forecast using the EnKF updating algorithm is introduced to have an algorithm-wise parallel comparison between EnKF and pure En3DVar.

In perfect-model OSSEs, DfEnKF and pure En3DVar perform differently when using the same localization radii. The serial (for DfEnKF) versus global (for pure En3DVar) nature of algorithms, and direct filter update (for DfEnKF) versus variational minimization (En3DVar) in nonlinear reflectivity observation operator are the major reasons for the differences. When the algorithms are tuned optimally, hybrid En3DVar does not outperform EnKF and pure En3DVar in perfect-model OSSEs, though their analyses are all much better than 3DVar. Under the perfect-model assumption, when ensemble background error covariance is a good estimation of the true error distribution, pure ensemble-based DA methods can do a good job, and the advantage of static background error covariance is not obvious in hybrid DA. In imperfect-model OSSEs, model errors are introduced by using different microphysical schemes in the truth run (Lin scheme) and the forecasts (WSM6 scheme) within the DA cycles. Hybrid En3DVar then outperforms 3DVar (and EnKF, DfEnKF, and pure En3DVar) for better capturing the intensity of the storm (the hail analyses below the freezing level) in the analysis. The advantage of hybrid En3DVar over pure ensemble-based methods is most obvious when ensemble background errors are systematically underestimated. For the real case, hybrid En3DVar outperforms 3DVar (EnKF and DfEnKF) in better capturing the hook echo structure and the rotating updraft (intensity) of the main tornadic supercell. The low-level mesocyclone is better forecasted when using hybrid En3DVar than other methods, suggesting stronger rotation and a larger tornado threat in the forecast.