10:45 〜 11:00
[MGI26-06] Combining Data Assimilation and Data-driven Sparse Sensing Placement Method For Designing Better Observation Locations
キーワード：データ同化、スパースセンサ最適化、観測位置
Data assimilation (DA) plays an important role in numerical weather prediction (NWP) to provide optimal initial conditions by combining forecasted state and observation data. There have been various DA studies to evaluate impacts of assimilated observations such as by ensemble forecast sensitivity to observation (EFSO). Our previous EFSO study found that coastal radiosonde observations surrounding the East China Sea were important for improving severe precipitation forecasts in Japan in 2018 partially because of relatively sparse observations over the ocean. For accurately estimating initial conditions over the ocean, the effective use of mobile radiosonde observations by aircraft and ships would be useful. However, there have been few studies yet that try to optimize the placement of mobile observations for NWP.
This study aims at designing better observation networks using the data-driven sparse sensor placement (SSP) method explored in informatics science. This method determines the optimal sensor locations so that the selected sensors effectively determine coefficients of proper orthogonal decomposition (POD) modes. The original SSP method reconstructs the spatial patterns of data from the selected sensors by solving a linear inverse problem using the POD modes. This study combined the SSP and DA so that we can accurately estimate the spatial patterns owing to Tikhonov regularization.
We applied the combined approach to two problems: statics and mobile observations. Firstly, the proposed method was applied for the placements of rain-gauge observations over Hokkaido Island in Japan. The optimized rain-gauge locations by the SSP reconstruct more accurate spatial patterns of precipitation than the fields reconstructed by operational stations known as AMeDAS. The second problem aims to optimize the locations of additional mobile stations for NWP. We implemented the SSP into an intermediate global atmospheric model coupled with the local ensemble Kalman filter (a.k.a. SPEEDY-LETKF) to optimize observing placement over the ocean. Our preliminary experiment was promising, showing that the SSP-based placement provides more accurate analyses than an ensemble spread-based placement.
This study aims at designing better observation networks using the data-driven sparse sensor placement (SSP) method explored in informatics science. This method determines the optimal sensor locations so that the selected sensors effectively determine coefficients of proper orthogonal decomposition (POD) modes. The original SSP method reconstructs the spatial patterns of data from the selected sensors by solving a linear inverse problem using the POD modes. This study combined the SSP and DA so that we can accurately estimate the spatial patterns owing to Tikhonov regularization.
We applied the combined approach to two problems: statics and mobile observations. Firstly, the proposed method was applied for the placements of rain-gauge observations over Hokkaido Island in Japan. The optimized rain-gauge locations by the SSP reconstruct more accurate spatial patterns of precipitation than the fields reconstructed by operational stations known as AMeDAS. The second problem aims to optimize the locations of additional mobile stations for NWP. We implemented the SSP into an intermediate global atmospheric model coupled with the local ensemble Kalman filter (a.k.a. SPEEDY-LETKF) to optimize observing placement over the ocean. Our preliminary experiment was promising, showing that the SSP-based placement provides more accurate analyses than an ensemble spread-based placement.