Abstract:As one of the essential variables in the ocean, sea surface temperature (SST) significantly impacts global climate and marine ecology, so it is necessary to forecast the sea surface temperature. Deep learning is highly efficient at data processing, but it is rarely used in equatorial Pacific SST short-term forecasting. Based on Long Short-term Memory (LSTM) network, this paper constructs a daily forecast model of SST in the tropical Pacific Ocean (10°S-10°N, 120.0°E-280°E) in the next ten days. Using observations from 1982-2010 as a train set and data from 2011-2020 as preliminary values, the model forecasts the SST. The results show that the forecast Root Mean Square Error (RMSE) in the eastern equatorial Pacific region is more significant than that in the central and western areas. The RMSE of the east part is about 0.6°C on the first day of the forecast, while the west and central regions are less than 0.3°C. In different interannual variation phases, RMSE is the largest in the La Ni?a period, followed by normal years, and the smallest in the El Ni?o period. RMSE in the La Ni?a period is more than 20% in some regions than in the El Ni?o period. The forecast deviation is positive in the east and negative in the west. Regarding relevant forecasting techniques, the central part is the best. The number of predictable days is more than ten days. Specifically, the number of predictable days near the equatorial cold tongue is 4-7 days, and the number of predictable days in parts of the western equator is three days. Forecasting models generally have lower monthly fore-casting skills than the western regions in the eastern equatorial Pacific region, and the areas performed the worst in October and November. In general, the SST forecast model based on LSTM can well capture the evolution characteristics of SST in time series, and the forecast performance is good in different cases. At the same time, the forecast model relying on data-driven can quickly and well predict the short-term change of daily average SST within the next ten days.