Reconstruction of missing data in geophysical fields. resolution of moving patterns

A method for the reconstruction of missing data in large data sets is presented. The method, DINEOF (Data INterpolating Empirical Orthogonal Functions), calculates the missing data from an optimal number of EOFs determined by cross-validation. A Lanczos method has been used for the EOF decomposition, in order to work with large matrices. DINEOF has been applied to two data sets of sea surface temperature: a set of 105 images in the Adriatic Sea and a set of 216 images in Tanganyika Lake. These data sets present 52% and 37% of missing data respectively, due to cloud coverage. Several validation studies have been carried out: comparison with in situ data and reconstruction of increasing amounts of missing data, from 40% to 80% of the total data, by artificially adding clouds. All tests show that results are robust. DINEOF uses a classical EOF decomposition. In this work we also present a different EOF decomposition, known as Extended EOF (ExEOF) or Singular Spectrum Analysis (SSA). This technique consists in using a lagged version of the matrix being analysed. By taking into account both the spatial and temporal correlation of the data, the ExEOF technique resolves spatio-temporal moving patterns in a more accurate way. Preliminary results show that this technique helps to better reconstruct the missing data in our data sets