It is very important for model or function to have exact input parameters. However

the input parameters often have some error in practical application. The parameter error will lead to larger model error in prediction of non-linear model (parameter). Ensemble Kalman filter is introduced into non-linear model (parameter) for parameter estimation. And joint state vector is used to update model parameter and state in the same assimilation time. The method can dynamically adjust model parameters and states according to changing environment by assimilating observation data. And more importantly

it can release the accumulated model error. The method is applied to two-dimension non-linear model which changes with time step. Studies show that the method can obtain ideal results in parameter estimation. It also has good performance in robustness and self-adaption.