11:15 〜 11:30
[MGI26-08] Investigation of a Multiscale Variational Data Assimilation based on Wavelet
キーワード：データ同化、ウェーブレット、ハイブリッド四次元変分法
High impact weather events are often accompanied with localized and rapid changes in the atmosphere, which are at small scale in time and space. At the same time, these events are associated with large-scale environments, such as low-level warm moisture inflow and upper-level cold air. Thus, it is important to accurately predict atmospheric phenomena at various scales in forecasting these events. High-frequency and high-density observations have been made available due to rapid progress in remote sensing technology, which provides the information of the atmosphere involving severe weather events at high resolution in time and space. These remote sensing observations often have a wide coverage, indicating that they contain large scale information at the same time. Therefore, it is important to extract observational information covering a wide range of scales in data assimilation (DA) to effectively incorporate it to numerical weather prediction for higher forecast accuracy.
This study investigates an enhancement of the hybrid four-dimensional variational (4D-Var) DA scheme to take into account the multiscale aspect of the flow-dependent background error covariance. The investigation is based on a hybrid 4D-Var system of Fujita et al. (2022), which is an extension of the JMA nonhydrostatic model-based variational DA (JNoVA; Honda et al. 2005) 4D-Var that has been operated in the former Meso-scale analysis (MA; JMA 2019) of Japan Meteorological Agency (JMA) until 2020. The control variables of the hybrid 4D-Var assigned to the flow-dependent background error are further enhanced, and are constructed in a wavelet space.
The flow-dependent background error for each scale component shows that its spatial structure and inter-scale correlations vary in response to the atmospheric phenomena predominant at the scale. Because the distance where sampling error dominates in the ensemble background error correlation increases with scale, the localization for each scale is specified accordingly. This hybrid 4D-Var with the wavelet control variable is shown to eliminate sampling errors in increments at far ranges from the observation positions. The scheme provides spectral characteristics of the increments closer to that of a large ensemble size, compared to the hybrid 4D-Var with the conventional grid control variable. The wavelet control variable also shows a higher sensitivity to the scale properties of observations, such as a thinning interval, compared to the grid control variable. A cycling experiment is performed in which all observations used in MA and temporal and spatial high-density atmospheric motion vectors are assimilated. The first guess accuracy is more improved than those obtained by the 4D-Var and hybrid 4D-Var with the conventional grid control variable. Promising results are obtained by the verification of forecasts using radiosonde and observed precipitation.
References
Y. Honda, et al., 2005: Quart. J. Roy. Meteor. Soc., 131, 3465-3475.
T. Fujita, et al., 2022: Mon. Wea. Rev., 150, 481-503.
JMA, 2019: Outline of the operational numerical weather prediction at the JMA. JMA, Tokyo, Japan.
This study investigates an enhancement of the hybrid four-dimensional variational (4D-Var) DA scheme to take into account the multiscale aspect of the flow-dependent background error covariance. The investigation is based on a hybrid 4D-Var system of Fujita et al. (2022), which is an extension of the JMA nonhydrostatic model-based variational DA (JNoVA; Honda et al. 2005) 4D-Var that has been operated in the former Meso-scale analysis (MA; JMA 2019) of Japan Meteorological Agency (JMA) until 2020. The control variables of the hybrid 4D-Var assigned to the flow-dependent background error are further enhanced, and are constructed in a wavelet space.
The flow-dependent background error for each scale component shows that its spatial structure and inter-scale correlations vary in response to the atmospheric phenomena predominant at the scale. Because the distance where sampling error dominates in the ensemble background error correlation increases with scale, the localization for each scale is specified accordingly. This hybrid 4D-Var with the wavelet control variable is shown to eliminate sampling errors in increments at far ranges from the observation positions. The scheme provides spectral characteristics of the increments closer to that of a large ensemble size, compared to the hybrid 4D-Var with the conventional grid control variable. The wavelet control variable also shows a higher sensitivity to the scale properties of observations, such as a thinning interval, compared to the grid control variable. A cycling experiment is performed in which all observations used in MA and temporal and spatial high-density atmospheric motion vectors are assimilated. The first guess accuracy is more improved than those obtained by the 4D-Var and hybrid 4D-Var with the conventional grid control variable. Promising results are obtained by the verification of forecasts using radiosonde and observed precipitation.
References
Y. Honda, et al., 2005: Quart. J. Roy. Meteor. Soc., 131, 3465-3475.
T. Fujita, et al., 2022: Mon. Wea. Rev., 150, 481-503.
JMA, 2019: Outline of the operational numerical weather prediction at the JMA. JMA, Tokyo, Japan.