A new tool for spatiotemporal pattern decomposition based on empirical mode decomposition: A case study of monthly mean precipitation in Taihu Lake Basin, China

Abstract

We present a new tool for spatiotemporal pattern decomposition and utilize this new tool to decompose spatiotemporal patterns of monthly mean precipitation from January 1957 to May 2015 in Taihu Lake Basin, China. Our goal is to show that this new tool can mine more hidden information than empirical orthogonal function (EOF). First, based on EOF and empirical mode decomposition (EMD), the time series which is an average over the study region is decomposed into a variety of intrinsic mode functions (IMFs) and a residue by means of EMD. Then, these IMFs are supposed to be explanatory variables and a time series of precipitation in every station is considered as a dependent variable. Next, a linear multivariate regression equation is derived and corresponding coefficients are estimated. These estimated coefficients are physically interpreted as spatial coefficients and their physical meaning is an orthogonal projection between IMF and a precipitation time series in every station. Spatial patterns are presented depending on spatial coefficients. The spatiotemporal patterns include temporal patterns and spatial patterns at various timescales. Temporal pattern is obtained by means of EMD. Based on this temporal pattern, spatial patterns at various timescales will be gotten. The proposed tool has been applied in decomposition of spatiotemporal pattern of monthly mean precipitation in Taihu Lake Basin, China. Since spatial patterns are associated with intrinsic frequency, the new and individual spatial patterns are detected and explained physically. Our analysis shows that this new tool is reliable and applicable for geophysical data in the presence of nonstationarity and long-range correlation and can handle nonstationary spatiotemporal series and has the capacity to extract more hidden time-frequency information on spatiotemporal patterns.