Abstract. A data-driven method for extracting temporally persistent information, at different spatial scales, from rainfall data (as measured by radar/satellite) is described, which extends the Empirical Mode Decomposition (EMD) algorithm into two dimensions. The EMD technique is used here to decompose spatial rainfall data into a sequence of high through to low frequency components. This process is equivalent to the application of successive low-pass spatial filters, but based on the observed properties of the data rather than the predetermined basis functions used in traditional Fourier or Wavelet decompositions. It has been suggested in the literature that the lower frequency components (those with large spatial extent) of spatial rainfall data exhibit greater temporal persistence than the higher frequency ones. This idea is explored here in the context of Empirical Mode Decomposition. The paper focuses on the implementation and development of the two-dimensional extension to the EMD algorithm and it's application to radar rainfall data, as well as examining temporal persistence in the data at different spatial scales.