In GNSS applications, carrier-smoothed-code is a widely used technique to combine code pseudo-range and carrier phase measurements. A dynamical ionospheric delay modeling method is proposed based on Kalman filter and least-squares theory. The level of the process noise is adaptively tuned along with the real-time KF state estimation, based on the online variance component estimation method. Meanwhile, the correlations of the time differenced carrier phase measurements are considered. This approach avoids overly optimistically evaluating the estimate and improves the transient accuracy of the estimates. A real GPS dataset is employed to check the performance of the proposed method under different conditions. The results show that the new algorithm can model the ionospheric delay variation well with different sampling intervals or even in ionospheric abnormal environment. The positioning accuracy can be confirmed, about 21%, 35% and 16% better are obtained in the N, E, and U direction than raw dataset.