The process of gathering of scientific data plays an important role in telemonitoring and communications technologies of underwater information. However, to obtain such a huge data in underwater-wireless-sensor-networks, conventional methods often fail in energy efficiency. Compressive sampling (CS) provides a new perspective to solve the problem. Unfortunately, the underwater acoustic signal is non-sparse in the time domain and the current CS methods cannot be used directly. This paper adopts the discrete cosine transform-based dictionary-matrix for sparse representation. In addition, the measurement matrix is optimized via the steepest descent method for more efficient sampling. Then, we introduce an approach based on an approximated l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> norm at the receiving terminal, to search the sparse solution via the steepest descent method and projections. Combing with the measurement matrix and dictionary-matrix, the sparse estimation is used for reconstruction. Experimental results confirm the superior performances of the strategies of the proposed compressive sampling and reconstruction than those of the traditional measurement matrix and reconstruction methods, including matching pursuit and orthogonal matching pursuit methods.