<p>Remote sensing imagery plays a vital role in applications such as disaster assessment and resource exploration. However, it is often degraded by atmospheric conditions such as haze and clouds in practical acquisition, resulting in reduced contrast, blurred details, and degraded visual quality, which hinder subsequent information extraction and analysis. To address this challenge, this paper proposes a hierarchical dynamic sampling Mamba for remote sensing image dehazing. The proposed approach integrates Mamba’s state space modeling capability into a multi-scale hierarchical framework and incorporates a dynamic sampling strategy to adaptively guide the state update process. This enables the network to focus more effectively on structurally salient regions and densely hazy areas. Furthermore, a cross-scale feature fusion scheme is employed to jointly optimize global context modeling and local detail restoration, allowing the network to accurately capture haze distribution patterns at different spatial scales. Experiments on the publicly available SateHaze1k and RICE remote sensing image degradation datasets show that the proposed method consistently outperforms state-of-the-art dehazing approaches in terms of PSNR and SSIM. Visual comparisons further confirm its superiority in detail recovery and texture fidelity.</p>