Abstract: Abstract： In order to solve the problem of the relationship between navigationobservability and accuracy in nonlinear navigation system, the observability analysis method based on the error covariance matrix is adopted.The nonlinear predictive filter(NPF) is improved combined with extended Kalman filter(EKF),and the error covariance of improved predictive filter is deduced. Then the eigenvalue decomposition is carried out to analyze the relationship between the eigenvalue and navigation accuracy.Taking the autonomous navigation system of the small celestial probe landing as an example, the simulation validation is carried out and the validity and accuracy of the improved NPF algorithm are verified comparing with the EKF.The influence of different error factors on observability is analyzed, which provide a reference for the filter design of autonomous navigation system in the actual spacecraft process.