In this paper we explore the performance of a selective decision directed channel estimation (CE) method for underwater acoustic (UWA) orthogonal frequency division multiplexing (OFDM) based communications. The method is based on hard or soft selection of data sub-carriers that are used in the decision directed mode. Particularly, in the process of hard selection we choose those sub-carriers that have high signal-to-noise ratios (SNR), while in the process of soft selection we weight the data sub-carriers for reliable channel estimation. The latter approach is shown to be advantageous since the channel state information (CSI) from all data sub-carriers is exploited. We couple selective decision directed CE with different channel reconstruction methods such as least-squares (LS) with thresholding, orthogonal matching pursuit (OMP), and basis pursuit denoising (BPD). Our results indicate good performance improvements as compared to the block-by-block pilot-assisted CE with a uniform pilot grid for high spectral efficiency communications. Moreover, we explore the trade-offs resulting from using a varying number of pilots and/or decision directed data sub-carriers in terms of the overall system throughput and bit error rate (BER). Experimental results are obtained using real data measurements from the Mobile Acoustic Communication Experiment 2010 (MACE'10), conducted off the coast of Martha's Vineyard.