The Kalman filter these two sources by assigning a confidence (called covariance ) to each. If the GPS is very noisy, the filter trusts the model more, and vice versa.