The limited sample size of Sea Surface Temperature (SST) data on large time scales (e.g.,weekly or monthly) poses significant challenges,including weak model generalization,low prediction accuracy,and high model complexity.To address these issues,this paper proposes an intelligent prediction method for large-scale SST that integrates transfer learning and knowledge distillation.First,a source-domain SST prediction model is established by combining Graph Convolutional Neural Networks (GCNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks to fully extract spatio-temporal features from daily-scale SST data.A Generative Adversarial Network (GAN) is then employed for feature alignment to facilitate domain adaptation,aligning the feature distributions of the source and target domains to enhance the model's predictive performance in the target domain.Finally,a multi-task learning-based knowledge distillation network is developed to mitigate the issues of high parameter redundancy and low computational efficiency in deep transferred networks.Experimental results demonstrate that,under various time-scale SST conditions,the proposed transfer learning approach achieves at least a 7.00% reduction in root mean square error compared to baseline models without this mechanism,effectively improving prediction accuracy in small-sample scenarios.Furthermore,the multi-task knowledge distillation network reduces the number of network parameters by 88.21% and 74.68% compared to the transferred deep network,thereby enhancing the timeliness of SST prediction.