Multidimensional Feature Explorer for Unbalanced Spatiotemporal Data

Feature analysis of weak nonlinear signals from geographic spatiotemporal data has received increasing attention. Most existing signal processing methods cannot effectively perform comprehensive feature analysis because of the multiple dimensions and unbalance of spatiotemporal data. We developed a divide-aggregate-explore method for such analysis of spatiotemporal data. In our method, different dimensions are first divided for multidimensional analysis, and the tensor-block structure is adopted to reorganize the original data and distinguish differences in dimensions. Then, information-based data aggregation is used to weaken the impact of dimensional unbalance. Case studies based on climatic reanalysis of field data released by the National Oceanic and Atmospheric Administration showed that the proposed method can effectively extract weak propagation signals such as the El Niño-Southern Oscillation (ENSO) and ENSO Modoki. Our method can also reveal more detailed evolutionary characteristics of complex coupling systems in different dimensions compared with classical feature detection methods such as principal component analysis and tensor decomposition.