MUSCAT: Multi-Scale Spatio-Temporal Learning with Application to Climate Modeling

• DOI: 10.24963/ijcai.2018/404
• Publication Date: 2018-07-05T05:49:10Z
• Authors: Jianpeng Xu; Xi Liu; Tyler Wilson; Pang-Ning Tan; Pouyan Hatami; Lifeng Luo
• Keywords: Temporal scales; Temporal database

In climate and environmental sciences, vast amount of spatio-temporal data have been generated at varying spatial resolutions from satellite observations computer models. Integrating such diverse sources has proven to be useful for building prediction models as the multi-scale may capture different aspects Earth system. this paper, we present a novel framework called MUSCAT predictive modeling multi-scale, data. performs joint decomposition multiple tensors scales, taking into account relationships between variables. The latent factors derived tensor are used train temporal scales each location. outputs these ensemble will aggregated generate future predictions. An incremental learning algorithm is also proposed handle massive size tensors. Experimental results on real-world United States Historical Climate Network (USHCN) showed that outperformed other competing methods in more than 70\% locations.