[M-GI29] Data assimilation: A fundamental approach in geosciences

[MGI29-07] From Predictability to Controllability: Control Simulation Experiment (CSE)

*三好 建正¹、Sun Qiwen^1,2、寺崎 康児¹、前島 康光¹ (1.理化学研究所、2.名古屋大学)

キーワード：制御シミュレーション実験、観測システムシミュレーション実験、数値天気予報、データ同化

The Observing Systems Simulation Experiment (OSSE) is a very powerful and widely applied approach to evaluate observing systems and data assimilation methods in numerical weather prediction (NWP). In the OSSE, we generate a nature run (NR) using a model and simulate observations by sampling the NR. An independent model run with data assimilation of the simulated observations mimics an NWP system, and we compare it with the NR to evaluate the observations and data assimilation method. In this study, we extend the OSSE and design the Control Simulation Experiment (CSE), in which we add perturbations to the NR and try to modify it to a desired state. Investigating what perturbations are effective to avoid a high-impact weather event would be useful to understand the controllability of such an event. Since the weather system is chaotic, and even more so for disturbances, small differences generally lead to big differences, particularly for high-impact weather events. This suggests potentially effective control, i.e., small interventions would lead to big differences for high-impact weather events. The chaos control has been studied extensively in the field of dynamical systems theory, but taking advantage of dynamical instability to avoid certain trajectories has not been a main focus to the best of the authors’ knowledge. We first tested this idea with the Lorenz-63 3-variable model and performed an OSSE with an ensemble Kalman filter (EnKF). We extended the OSSE by adding very small perturbations (only 3% of the observation error) to the NR and found an effective approach to control the trajectory to stay in one side of the Lorenz’s butterfly attractor without shifting to the other. Following the implications and understandings from the Lorenz-63 model experiments, we tested with the Lorenz-96 40-variable model to avoid the occurrences of extreme values, mimicking to avoid extreme events in NWP. Finally, we further extended the idea to test with realistic global and regional NWP models. This presentation will summarize the concept and methodology of CSE with some proof-of-concept demonstrations with the toy models and realistic NWP models. This is an attempt to a potential paradigm change of NWP research from decades of predictability to the new era of controllability.