The orthogonal time frequency space (OTFS) technique is an innovative modulation scheme that provides significant advantages in terms of channel delay and Doppler shifts. In this work, we study the sparse delay and Doppler channel estimation problem for OTFS and consider the impact of inheriting initial and iterative parameters on adjacent estimated channel corresponding to previous OTFS transmitted blocks. We propose a parameter-inherited sparse Bayesian learning (SBL) channel estimation algorithm based on unitary approximate message passing (UAMP). Simulation results show that compared to the state-of-art SBL-based algorithms, the proposed algorithm has faster convergence speed and higher accuracy. Furthermore, by exploiting the block circulant matrix with circulant blocks (BCCB) matrix property, we replace the matrix multiplication with two-dimensional (2D) fast Fourier transform (FFT), which leads to a low complexity.