Carbon Dioxide Emission Forecasting for Medium and Large Reservoirs in China
Reservoir CO2 emissions are a critical component of the global carbon cycle, yet existing prediction models often neglect reservoir size stratification, mixing data from large, medium, and small reservoirs. This study addresses this gap by constructing a training dataset exclusively for large and medium reservoirs in China and developing a neural network-based CO2 flux prediction model. The model revealed significant latitudinal zonation, with CO2 fluxes in southern low-latitude basins (e.g., Yangtze, Pearl) substantially higher than in northern high-latitude basins (e.g., Yellow, Songliao). Statistical analysis identified total phosphorus (TP) content as the strongest driver, exhibiting a positive exponential correlation with CO2 flux (p < 0.01). Reservoir age and latitude showed significant negative correlations, while physical parameters such as reservoir area, pre-impoundment submersion ratio, and average water depth showed no significant national-scale correlation. The study underscores the heterogeneity of dominant drivers across basins and proposes differentiated carbon mitigation strategies: for southern high-emission basins, controlling phosphorus inputs through wastewater treatment and restricting aquaculture; for northern basins, enhancing soil conservation to reduce carbon-rich sediment influx; and for new reservoirs, implementing vegetation clearance before impoundment and optimizing flood discharge schedules. These findings provide a theoretical basis for carbon emission characterization and support low-carbon management of large and medium reservoirs in China.