An Intelligent Reconstruction Method for Sea Surface Temperature Profile in Dynamic Sea Conditions Based on BP Neural Network

Sea Surface Temperature exerts a significant influence on marine dynamic processes, ocean-atmosphere change processes, and the material cycling within ecosystems. Traditional temperature observation methods suffer from insufficient vertical resolution for acquiring Sea Surface Temperature under dynamic sea conditions due to interference. Aiming at the difficulty in obtaining high-resolution Sea Surface Temperature data under such dynamic conditions, this paper innovatively proposes an intelligent reconstruction method for sea surface temperature profile in dynamic sea conditions based on back propagation neural network. Initially, centimeter-level raw data support is provided by an independently developed 48-measurement-point vertical temperature array with 1-cm vertical spacing. Subsequently, a multi-sensor collaborative method, utilizing a hybrid Back Propagation neural network model optimized by the Levenberg-Marquardt algorithm, fuses this raw temperature data with three-dimensional parameters (wave height, wind speed, buoy attitude) from the hydro-meteorological sensing system to predict high-precision sea surface skin temperature data. Finally, the Sea Surface Temperature profile is reconstructed using the Empirical Orthogonal Functions reconstruction method. In November 2024, a verification experiment was conducted in the northern part of the South China Sea, where the effectiveness and performance of this method were evaluated and verified. The results demonstrate a mean absolute error of 0.1416°C for predicted values, with a remarkably high correlation (98%) to true value measurements. This method provides critical data support for research in air-sea heat flux, extreme weather warnings, and evaporation duct height prediction.