The processes and phenomena hidden in massive marine data importantly impact heat transportation, material transportation, and climate formation. Visualization can assist people in mining and understanding marine data to gain insight. Thus, oceanographers must study ocean processes and phenomena. However, one remaining challenge in the existing visualization methods is efficiently rendering marine data with large volumes and illustrating the internal structure of marine phenomena. To solve this problem, we propose a new visual analytics framework involving 4 parts for visualizing extensive marine scalar volume data. We first use a single box and double spheres separately as proxy geometries to draw a flat Earth and spherical Earth. Second, we design a new ray-casting algorithm based on graphics processing unit to reduce the volume of marine data. This algorithm accelerates volume rendering by using an adaptive texture sampling rate and step size. Third, we use a depth correction algorithm to accurately restore the ocean scale. Finally, we develop an internal roaming algorithm to observe the internal structure of marine data. In this way, users can dynamically observe the internal structure of marine phenomena. To illustrate the effectiveness of our algorithms, we use them to visualize Hybrid Coordinate Ocean Model data and Argo data.