New Developments in Crowd Sourcing Tropical Cyclone Intensity Estimates

Cyclone Center is a crowd sourcing project developed to provide homogeneous estimates of historical tropical cyclone (TC) intensity and uncertainty from infrared imagery. Since its launch in September 2012, we’ve collected over 730,000 unique classifications from 14,500 users. Initial examination of Cyclone Center intensities exhibited low bias and mean error when compared to a “ground truth” set of storms that were observed with aircraft reconnaissance data.

The Cyclone Center algorithm, which relies heavily on the traditional Dvorak Enhanced Infrared (EIR) technique, has been recently enhanced by weighting individual classifiers through an Expectation Maximization (EM) algorithm. The EM technique is an iterative statistical process that weights user performance by comparing their classifications of scene type (e.g., eye storm, curved band, embedded center) to the most likely correct answer for a particular TC image. The classifications of users with the higher scores are then weighted more toward the final intensity estimate.

A presentation of the global tropical cyclone activity for 2005 - as determined by citizen scientists - will be made to highlight the performance of the enhanced Cyclone Center algorithm. The anticipated completion of the 32-year data set may provide a valuable resource for a global TC reanalysis.