The actual underwater acoustic channel has the feature of clustered sparsity. To solve this problem, a hierarchical Bayesian model for underwater acoustic channel estimation is proposed, which is based on Markov chain Monte Carlo (MCMC) sampler in orthogonal frequency division multiplexing (OFDM) underwater acoustic communication. The prior distribution of the channel is formulated by the structural features of the clustered sparsity underwater acoustic channel. The posterior distribution of the channel model parameters is achieved by Bayesian inference and the likelihood function of the received signal. Finally, the MCMC sampler is utilized to sample the posterior conditional distribution of the channel model parameters. In this way, the maximum posterior estimation of the clustered sparsity underwater acoustic channel can be obtained. The performance of the proposed method is analyzed by simulation comparisons in terms of least squares, matching pursuit and stepwise orthogonal matching pursuit channel estimation methods under different received signaltonoise ratios. The lake test shows that the proposed method can achieve accurate OFDM underwater acoustic channel estimation, tracking and decoding without any channel prior information. The method realizes OFDM underwater acoustic communication with the communication distance of 600 m to 3 500 m. In addition, the transmission data rate can reach 608 kbps.