The signal denoising problem is a major challenge in underwater communication (UWC). The denoising problem in UWC has been addressed, where channel estimation approaches are provided in terms of sparsity, time, and frequency. Convex optimization is currently widely used for recovering sparse signals from compressed data. A nonuniform sparse 2D frequency-domain channel recovery scheme in the delay-Doppler domain representation is provided as a mathematical framework to solve channel estimation problems. Compressive sensing (CS) at higher Doppler frequencies (high channel mobility) and near full-rate sampling at lower Doppler frequencies (stagnant channel) are used to recover both rapid-fluctuating as well as slow-varying channel components with high precision. The prior knowledge of modified basic-pursuit denoising channel estimation based on nonuniform CS and sparsity-constrained least-squares has been discovered, where <i>l</i><sub>1</sub>-minimization of element wise multiplication of the equal-size vectors is performed. The proposed channel estimation scheme is compared with conventional CS over a sampling ratio, ambient noise level, and observation window length. The acoustic channel simulations and experimental channel data from SPACE08 are used to verify prior knowledge provided by the proposed scheme. Simulation results confirm the superiority of the proposed algorithm for channel estimation over the existing ones.