Abstract: Coal mine water damage frequently results in severe casualties, significant property losses, and reduced recoverable mine reserves. It can also precipitate secondary disasters like ground subsidence and collapse, making accurate prediction and forecasting of coal mine water hazards a critical area of research for mine water damage prevention and control. Despite recent advancements in coal mine water hazard prediction technology, there remains an urgent need for new detection techniques to meet the requirements of safe, efficient, intelligent, and green mining operations for quantitative and precise water hazard assessments. Magnetic Resonance Sounding (MRS) is the only geophysical method capable of directly and quantitatively detecting water. However, the surface nuclear magnetic resonance method (SNMR) is limited by shallow detection depth, weak anti-interference capabilities, and poor delineation of water-bearing body boundaries. To address these limitations, the surface-tunnel nuclear magnetic resonance method (STNMR) is proposed, based on the SNMR and incorporating the practical construction methods of underground coal mines. This method involves placing a large-size transmitting loop on the surface and a multi-turn receiving loop underground. Firstly, the theoretical derivation of the STNMR method was conducted. Based on the conventional aquifer model, the response characteristics and the distribution law of the kernel function of the STNMR signal were analyzed. Through numerical computation, the theoretical signal strength and effective detection depth of the STNMR method and the SNMR method were compared and analyzed. The results indicated that in the aquifer with a burial depth exceeding 150 meters, the STNMR method exhibits a higher amplitude response signal, and its detection depth is nearly twice that of the SNMR method. Subsequently, numerical simulation calculations were performed for the water-bearing bodies of the roof of coal seam and the floor of coal seam, employing three data acquisition methods: synchronous data acquisition (SDA), fixed-receiver data acquisition (FRDA), and fixed-transmitter data acquisition (FTDA). The E₀ multichannel results were utilized to assess the capacity of different data acquisition methods in delineating the boundaries of the water-bearing bodies. The findings show that the SDA method has a slightly weaker ability to distinguish the center of the water body in the roof of coal seam and the boundary of the water body at the floor of coal seam. In contrast, both the FRDA and FTDA methods demonstrate good performance in identifying the center and boundaries of the water bodies in the roof and at the floor of coal seam in the lateral direction. Finally, the 1D inversion of free induction decay signals for various layered models and the proposed Quasi-2D inversion for the water body model of the roof and floor of coal seam were carried out using the 1D Occam's inversion algorithm. The ability of quantitative detection of water content in the rock layer by STNMR is verified.