Mangrove ecosystems can be both significant sources and sinks of greenhouse gases. The restoration of mangrove forests is increasingly used as a natural climate solution tool to mitigate climate change. However, the estimates of carbon exchanges remain unclear, especially from restored mangroves. In this study, we observed the temporal variations in carbon dioxide (CO2) and methane (CH4) fluxes and their biophysical controls for 4 years, based on a closed-path eddy covariance (EC) system. The measurements were conducted in a mangrove wetland park with 14-year-old restored mangroves surrounded by open waters in Guangdong Province, China. The EC measurements showed that the mangrove ecosystem acted as a CO2 source with a net CO2 ecosystem exchange (NEE) of 305 g C m−2 from January 2019 to May 2020 by the 5-m tower measurement. After the tower was adjusted to 10 m, the mangrove showed a CO2 sink with an NEE of −345 g C m−2 from June 2020 to December 2022. The change in tower height influenced the interpretation of interannual trends on NEE. There were no significant interannual trends in the gross primary productivity (GPP) and the ecosystem respiration (Re) values. The change from CO2 source to sink may be attributed to the decrease in land surface proportion by the tower replacement, which reduces the proportion of the mangrove canopy respiration and, therefore, captures lower CO2 fluxes from open waters. The restored mangroves indicated strong CH4 sources of 23.2–26.3 g C m−2 a−1. According to the random forest analysis, the land surface proportion, radiation, and relative humidity were the three most important predictors of NEE, while the CH4 flux was most sensitive to air temperature. Compared to the natural and long-term restored mangroves, this 14-year-old restored mangrove had not yet achieved a maximum carbon sequestration capability. Our study highlights the need for the careful design of long-term observations from restored mangroves and proposes future needs in the context of carbon neutrality.