A Periodic Convolutional Recurrent Network Model for Climate Prediction

Abstract

A prediction on spatiotemporal climate data that uses a recurrent network is aiming to predict future spatial data by learning from prior spatial sequence data. Most machine learning researches on this domain do not consider periodic patterns, which is essential for climate data. Inspired by Periodic-CRN, we propose a predictive model by using a convolutional long short-term memory as a convolutional recurrent network (CRN). The model also has the mechanisms that load and save a periodic representation and combined to current representation to improve the accuracy result.