We explore to what extent data-driven prediction models have skills in forecasting daily sea-surface temperature (SST), which are comparable to or perform better than current physics-based operational systems over long-range forecast horizons. Three hybrid deep learning-based models are developed within the South China Sea (SCS) basin by integrating deep neural networks (back propagation, long short-term memory, and gated recurrent unit) with traditional empirical orthogonal function analysis and empirical mode decomposition. Utilizing a 40-year (1982–2021) satellite-based daily SST time series on a 0.25° grid, we train these models on the first 32 years (1982–2013) of detrended SST anomaly (SSTA) data. Their predictive accuracies are then validated using data from 2014 and tested over the subsequent seven years (2015–2021). The models’ forecast skills are assessed using spatial anomaly correlation coefficient (ACC) and root-mean-square error (RMSE), with ACC proving to be a stricter metric. A forecast skill horizon, defined as the lead time before ACC drops below 0.6, is determined to be 50 days. The models are equally capable of achieving a basin-wide average ACC of ~0.62 and an RMSE of ~0.48 °C at this horizon, indicating a 36% improvement in RMSE over climatology. This implies that on average the forecast skill horizon for these models is beyond the available forecast length. Analysis of one model, the BP neural network, reveals a variable forecast skill horizon (5 to 50 days) for each individual day, showing that it can adapt to different time scales. This adaptability seems to be influenced by a number of mechanisms arising from the evident regional and global atmosphere–ocean coupling variations on time scales ranging from intraseasonal to decadal in the SSTA of the SCS basin.

In this paper, a novel and simple statistical prediction model for sea surface multivariate is developed based on extended empirical orthogonal functions (referred to as the MEEOF model). This simple model embeds the temporal evolution information into the empirical orthogonal function spatial patterns, which effectively captures the spatial distribution of the sea surface variables and their evolution process over time, and can be used to improve the accuracy of intra-seasonal ocean forecasts. At the same time, it considers both the correlation between different spatial and temporal points and the dynamic balance between different sea surface variables. The performance of the MEEOF prediction model has been examined in the South China Sea (SCS) based on remote sensing satellite datasets. Experimental results demonstrate that this model has significant forecasting capability within the forecast window of 15–90 days, compared with the traditional persistence forecasts and optimal climatic normal (OCN) results. Furthermore, this model exhibits good forecast performance throughout the entire forecast window; the final prediction model (referred to as the fusion model) is established by obtaining the weighted average for the MEEOF forecasts and persistence forecast results. Numerical experimental results show that this fusion prediction model consistently outperforms the persistence model, the OCN model, and the linear regression model over the entire forecast window. A case study shows that the propagation of ocean waves and the coordination between different sea surface variables can be well predicted by this simple model, indicating that this fusion model has a potential advantage in intra-seasonal predictions of the ocean.