Fig. 1. Model-driven versus data-driven data assimilation

Fig. 2. Comparison of the effects of SVR regression reconstruction on Lorenz63 trajectory with true state，Linear regression reconstruction of AnEnKS and model-driven assimilation of EnKS （Lorenz-63 X1 time series）

Fig. 3. RMSEs of the three methods with different M

Fig. 4. Variation of error e（t） in the assimilation process. The horizontal axis represents the time， and the vertical axis represents the error in the assimilation process

Fig. 5. Comparison of phase diagram trajectories between the two methods

Fig. 6. RMSE （SVR-DD-DA and AnEnKF） of two methods with different M and Ca values

Fig. 7. Assimilation results of the two methods with different M and OBS values （SVR-DD-DA and AnEnKF）

Fig. 8. Data assimilation performance of M = 100， N = 50， Ca = 0

Fig. 9. Data assimilation performance of M=10 000， N=50， Ca=0