Short-Gap Imputation of SST Time Series Using Statiscal and Deep Learning Techniques

Sea Surface temperature (SST) affects global climate and worldwide ecosystems. Through the analysis of the SST time series, events associated with the ocean and atmosphere can be understood and predicted. One of the main problems found in SST times series are missing values due to multiple factors. In this work, a comparative study between statistical and deep learning techniques was carried out and, an ensemble model is proposed based on the techniques that present the best performance for short-gap imputation. The techniques that were implemented include Autoregressive Integrated Moving Average (ARIMA), Spline Interpolation, Exponential Weighting Moving Average (EWMA), Linear Weighting Moving Average (LWMA), Long Short-Term Memory (LSTM), Bidirectional LSTM, Gated Recurrent Unit (GRU) and Bidirectional GRU (BiGRU). The results show that for short-gap imputation, statistical techniques outperform those based on deep learning, and the ensemble model based on statistical techniques outperforms standalone techniques in most cases.