Abstract: Thermal discharge from a nuclear power plant (NPP) can cause a sharp rise in local sea surface temperature (SST), exerting profound impacts on regional marine ecosystems. To address the limitations of traditional baseline temperature extraction methods, which are highly subjective and have limited regional applicability, this study proposes an adaptive reference temperature extraction method for thermal discharge from nuclear power plant thermal discharge: the cumulative histogram second-gradient method. Using Landsat 9 SST thermal infrared satellite data, this method was applied to two typical sea areas: the Fuding nuclear power plant (a semi-enclosed sea area) and the Fangchenggang nuclear power plant (an open sea area). First, the cumulative histogram of sea surface temperature (SST) in the discharge area was fitted using a generalized logistic function. Then, based on the spatial characteristics of the temperature-rise distribution of the thermal discharge, the reference temperature was automatically determined from the second-order derivative curve. The results show that the reference temperatures extracted by the cumulative histogram second-gradient method are in good agreement with those obtained by the bay average method (suitable for semi-enclosed sea areas) and the discrete multi-point average method (suitable for open sea areas), as recommended in the Technical specification to thermal discharge monitoring for coastal nuclear power plants based on satellite remote sensing(on trial). The absolute differences are ≤0.15 ℃, with mean differences of 0.12 ℃ (for Fuqing NPP) and 0.05 ℃ (for Fangchenggang NPP). Compared with conventional methods, the proposed method requires no manual threshold setting or selection of temperature reference areas, effectively overcoming their limitations and providing more convenient and reliable technical support for remote sensing monitoring of thermal discharge from NPPs.