Nonlinear (MARS) modeling of long-term variations of surface UV-B radiation as revealed from the analysis of Belsk, Poland data for the period 1976?2000

• Main Author: Krzy?cin, J. W.
• Other Authors: Institute of Geophysics
• Format: Article in Journal/Newspaper
• Language: English
• Published: HAL CCSD 2003
• Subjects: [SDU.OCEAN]Sciences of the Universe [physics]/Ocean; Atmosphere; [SDU.STU]Sciences of the Universe [physics]/Earth Sciences; North Atlantic
• Online Access: https://hal.science/hal-00317151; https://hal.science/hal-00317151/document; https://hal.science/hal-00317151/file/angeo-21-1887-2003.pdf

International audience A new, powerful statistical technique, multivariate adaptive regression splines (MARS), is applied to reproduce monthly fractional deviations of UV-B doses over Belsk, Poland, during the snowless (May?October) part of the year in the period 1976?2000. Two kinds of regressors were used: local ones (total ozone, percentage of sky covered by low-, mid-, high-level clouds or total solar radiation over Belsk) and non-local ones, i.e. those describing the long-distance forcings on the surface UV-B due to changes in the global atmospheric circulation. Standard indices of the Quasi-Biennial, North Atlantic, El Niño-Southern Oscillations, and the 11-year solar activity were used as non-local regressors. The results there indicate that the MARS procedure is able to reproduce the observed year-to-year and decadal oscillations in the UV data. The MARS model yields better model-observation agreement than an ordinary least-squares fit based on the same set of regressors. It is found that MARS is capable of handling interactions between the local and non-local regressors, suggesting a possible nonlinear nature of connections between variables characterizing the atmospheric transparency over Belsk and the long-distance forcings. MARS enables a reconstruction of the surface UV-B variations over any site based on the cloud and ozone data presently stored on web pages.