Abstract
Abstract. Empirical Mode Decomposition (EMD) is applied here in two dimensions over the sphere to demonstrate its potential as a data-adaptive method of separating the different scales of spatial variability in a geophysical (climatological/meteorological) field. After a brief description of the basics of the EMD in 1 then 2 dimensions, the principles of its application on the sphere are explained, in particular via the use of a zonal equal area partitioning. EMD is first applied to an artificial dataset, demonstrating its capability in extracting the different (known) scales embedded in the field. The decomposition is then applied to a global mean surface temperature dataset, and we show qualitatively that it extracts successively larger scales of temperature variations related, for example, to topographic and large-scale, solar radiation forcing. We propose that EMD can be used as a global data-adaptive filter, which will be useful in analysing geophysical phenomena that arise as the result of forcings at multiple spatial scales.
Highlights
Variability in the climate system occurs at a large range of space and time scales
The results presented above confirm that the 3 first Implicit Mode Surfaces (IMSs) qualitatively represent three different and hierarchical scales of the spatial organization of the global mean temperature: local to regional scale variations linked to the topography and the presence of ice-caps, the prominent warm pool region in the pacific, and the large-scale gradient from the tropics to the poles related to mean solar forcing
Empirical Mode Decomposition has been applied to the analysis of a two-dimensional field on the sphere, taking advantage of the data-adaptive nature of the method and the absence of edge or end effects on the sphere
Summary
Variability in the climate system occurs at a large range of space and time scales. One challenge of climate data analysis is to separate these scales and their interactions. The basic idea of EMD is to allow for an adaptive and unsupervised representation of the basic components of linear and non-linear signals and is designed to accommodate non-stationarity in the series This method has recently been extended to two dimensions (Linderheld, 2002; Nunes et al, 2003). On the sphere, it is advantageous to dispense with a formal coordinate system and to compute the distances between points of interest (extrema in the decomposition process), as one would do in Kriging a field of randomly arranged points in 2-D It is precisely this thinking that allows us to treat each observation with the same weight, in contrast with methods applied to the lat-long grid which cannot sensibly ascribe the same weight to values at high as well as low latitudes.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
