Modern approach to study trends in global ultraviolet index maps variation over solar cycles by the complementary use of time series, mathematical metrics and artificial neural networks

The feed forward, multilayered supervised artificial neural network dedicated to the reconstruction of world maps of erythemal Ultraviolet index (UV index) was constructed and trained using three different sets of data selected from 1978 to 2005. A combined analysis made using Manhattan, Euclidean, Bhattacharyya, Hellinger and topological Pompeiu-Hausdorff metrics has revealed large unconnected and inhomogeneous areas of high variability in UVI maps. The modeling of the long term variation in UVI during the whole sun activity cycle helped to understand trends and predict the direction of future variations in UV radiation level. The quality of prediction was evaluated using a newly proposed approach based on Pompeiu-Hausdorff metric.