As for the problem of quality evaluation method of sea surface temperature (SST) observed by satellite remote sensing. An analysis model of SST is proposed based on the combination of observations at different time, different places and with different techniques. According to this model, Kalman Filter and the principle of non-negative matrix factorization can be used to fuse the SST data in temporal and spatial dimension when the data absence occurs. Through which an accurate estimation of SST observations will be made. The experiment results with SST data obtained in East China Sea in 2006, showed that the model presented in this article can obviously improve the precision of SST data estimation, which can provide accurate reference for the quality evaluation of marine information.