Impact of RTPS and Radar Observation-Based Covariance Inflation Schemes on an Operational Convective-Scale Data Assimilation System over Taiwan
Journal
Weather and Forecasting
Journal Volume
40
Journal Issue
10
Start Page
2159
End Page
2177
ISSN
0882-8156
1520-0434
Date Issued
2025-10
Author(s)
Tsai, Chin-Cheng
Lien, Guo-Yuan
Schwartz, Craig S.
Jiang, Siou-Ying
Chang, Pao-Liang
Hong, Jing-Shan
Abstract
This study investigates a covariance inflation strategy combining the relaxation-to-prior spread (RTPS) and a radar observation-based random additive noise (RAN) scheme for an operational convective-scale local ensemble transform Kalman filter (LETKF) data assimilation (DA) system in the Central Weather Administration (CWA) of Taiwan. Continuously cycled DA experiments are conducted over two long periods: one for a case of 8-day consecutive afternoon thunderstorms and the other for a mei-yu frontal case over 2 days. Statistics of the prior (background) and posterior (analysis) ensemble means against radar reflectivity and radial velocity for both cases reveal that only applying the RTPS scheme is insufficient to attain a suitable ensemble spread and that the combined use of the RTPS and RAN schemes can much better improve the ensemble spread. Furthermore, quantitative precipitation forecasts are substantially improved when using both RTPS and RAN compared to only RTPS. Moreover, the sensitivity of different choices of perturbed variables in the RAN scheme is investigated. An evaluation considering forecast performance and model balance suggests that perturbing fewer variables (e.g., only the water vapor) in the RAN scheme may be recommended. SIGNIFICANCE STATEMENT: This study evaluates two covariance inflation schemes (relaxation-to-prior spread and radar observation-based random additive noise schemes) based on an operational ensemble data assimilation system. The results show that under long continuous cycles, the relaxation-to-prior spread combined with a random additive noise scheme can improve the background ensemble spread, analysis fields, and short-term quantitative precipitation forecasts. In addition, in the random additive noise scheme, adding random noises to only thermodynamic variables can achieve a spread expansion effect comparable to perturbing all four dynamic and thermodynamic variables while resulting in more balanced model initial conditions.
Subjects
Data assimilation
Numerical weather prediction/forecasting
Regional models
Publisher
American Meteorological Society
Type
journal article