A data mining algorithm for climate data: application to double ITCZ

A data mining algorithm is developed to analyze images collected over the Indian Ocean basin. Outgoing long-wave radiation (OLR) data are examined using image recognition techniques, in order to study inter-tropical convergence zone (ITCZ) anomalies in a region where they have previously received little attention. We examine the presence, location, and organization of minima in the data to detect and catalogue manifestations of the double (D)ITCZ. This is done via wavelet transform and multi-resolution analysis, in conjunction with a graph-based grouping algorithm. This process is calibrated using a heuristics based algorithm, and accuracy is examined using random sub-sampling of images. Finally, the algorithm is applied to NOAA's thirty-year reanalysis data to generate a DITCZ index for this region. We examine products of this index, including the seasonal variations in DITCZ presence. We also comment on the possibility of generalizing this technique to other data sets and other regions, as well as its potential application to studying the DITCZ over-prediction problem common to many global climate models.