Daily synoptic weather types of southern Indian Ocean: January 1979-October 2018

Each day is assigned to a 'winning' node between 1-9. e.g. 19790101 is assigned to SOM3. SOM code: "Kohonen" Package in R cran.r-project.org/web/packages/kohonen/index.html Study domain: 30°-75°S, 40°-180° E Time step / period: daily / January 1979-October 2018 Input data: ERA-Interim (apps.ecmwf.int/datasets/) 500hPa geopotential height (z500) daily anomalies. Climate Data Operators (CDO) was used to calculate the daily anomaly (code.mpimet.mpg.de/projects/cdo/) SOM algorithm parameters: refer to kohonen documentation for more information. dist.fcts (performance evaluation distance) = euclidean, grid = rect, neighbourhood function = gaussian, Nodes (number of SOM nodes) = 9, rlen (number of iterations) = 1000, alp (learning rate) = 0.05 to 0.01 and rad (radius) = 4 to 0.

In the Kohonen R-package, a radius value less than or equal to one corresponds to the point where only the 'winning' node is updated by each iteration, making the SOM algorithm similar to clustering techniques (e.g. k-means). The SOM algorithm used in this study was a hybrid between SOM/clustering (75% SOM, 25% clustering).