South China has the most abundant precipitation in the country. Current insufficient understanding of the mechanisms behind the complex decadal variability of South China summer monsoon rainfall (SCSMR) makes it challenging to skillfully predict such variability several years in advance. This study investigates the physical mechanisms and predictability of SCSMR from the perspective of upper-ocean heat content (OHC). Results show that upper-ocean heat content anomalies over the equatorial central-eastern Pacific provide a significant precursor signal, leading SCSMR by approximately 11 years. Low-level circulation anomalies over the North Pacific associated with central-eastern Pacific OHC excite westward-propagating oceanic Rossby waves through wind stress forcing, thereby modulating off-equatorial thermocline depth and sea surface temperature in the Pacific. These oceanic adjustments further regulate the western North Pacific anomalous anticyclone through Walker and Hadley circulation changes. The anomalous anticyclone subsequently modulates SCSMR through dynamic vertical moisture advection associated with ascending motion along the northwestern flank of the western Pacific subtropical high. Based on this physical mechanism, a two-layer long short-term memory (TL-LSTM) model is developed for decadal prediction. Compared with CMIP6 dynamical models, the model shows substantially higher skill, with a temporal correlation coefficient of 0.89 and mean square skill score of 0.72 during the independent hindcast period. These findings may serve as a reference for the decadal prediction of summer precipitation over South China, which strongly affects regional water resource management and infrastructure planning, poses severe challenges to disaster prevention and mitigation, as well as to socioeconomic sustainability.