Long‐term erythemal ultraviolet radiation in Novi Sad (Serbia) reconstructed by neural network modelling

Abstract

In this article, we proposed a simple neural network (NN) technique for estimating erythemal ultraviolet (UV) radiation in Novi Sad (Serbia) using available input parameters. The technique implies the use of one of two models depending on the availability of the input parameter: (a) NN model 1 (NNM1) which uses global solar radiation, clearness index, cloudiness and air mass; and (b) NN model 2 (NNM2) which adds total ozone content (TOC) to the NNM1 inputs. The three feed‐forward NNs with different internal structures and back propagation learning method for each NN model were used in modelling. The parallel calculation was used for learning each NN. The results showed that the NNM1 provides satisfactory estimate (R = 0.975, MBE = −0.614%, MAPE = 12.580%, RMSE = 17.716%) and that additional use of TOC NNM2 considerably improves the results (R = 0.982, MBE = −0.726%, MAPE = 10.161%, RMSE = 14.509%). The performances of developed NNMs become significantly better if warm part of the year is isolated (MAPE = 10.981 and 8.958; RMSE = 13.889 and 11.709, for NNM1 and NNM2, respectively). Variations of reconstructed annual averages of daily doses in the period 1949–2012 indicate ability of the technique to model the relationship between erythemal UV radiation and the affecting atmospheric factors. The analysis showed that the increasing trend during the warm part of the year in the period 1981–1996 was mainly caused by TOC, while the increase after 1996 was to a greater extent caused by cloudiness.

The importance of ground-based and satellite observations for monitoring and estimation of UV radiation in Novi Sad (Serbia)

“…(G, k, m, C, TOC) 0.998 -0.84 3.41 4.74 0.996 -0.84 3.45 4.24 Note. Adapted from "Long-term erythemal ultraviolet radiation in Novi Sad (Serbia) reconstructed by neural network modeling," by S. Malinovic-Milicevic, Y. Vyklyuk, M. M. Radovanovic, andM. D. Petrovic, 2018, International Journal of Climatology, 38(8), p. 3269.…”

“…The longest series of reconstructed UV radiation, from 1949 to 2012, was obtained using two neural network techniques depending on the availability of the predictors (Malinovic-Milicevic et al, 2018). Table 1 presents statistical parameters for the testing data set (2009)(2010)(2011)(2012) for both used models, separately for the whole year and in the warm part of the year (April-September).…”

“…Annual time-series of erythemal UV radiation (UVery), total ozone column (TOC) (a), and global solar radiation (G), clearness index (k) and cloudiness (C) (b) and their polynomial trends in Novi Sad for the period 1949-2012. Adapted from "Long-term erythemal ultraviolet radiation in Novi Sad (Serbia) reconstructed by neural network modeling," by S Malinovic-Milicevic, Y. Vyklyuk, M. M. Radovanovic, and. M. D. Petrovic, 2018, International Journal of Climatology, 38(8), p. 3269.…”

Climatological aspects of the increase of the skin cancer (melanoma) incidence rate in Europe

“…In Poland, Norway and the UK, the trend is~41 to~48% per 10 years and in Belgium~69% for women, while for men it is 48-51% in Poland and Norway, and 69-77% in the UK and Belgium (Figure 1). A common opinion is that such growth should be connected with a positive trend in the ambient EUV (Malinovic-Milicevic et al, 2018), the increasing use of sunbeds (Suppa et al, 2019), changes in human behaviour, and styles of clothing since twentieth century (Chang et al, 2014).…”

Reconstruction and variability of high daily erythemal ultraviolet doses and relationship with total ozone, cloud cover, and albedo in Novi Sad (Serbia)

“…In Novi Sad, two techniques have been applied to reconstruct daily UV ery doses. The first technique is based on radiation transfer in combination with the empirical method (Malinovic‐Milicevic et al ., 2015; Malinović‐Milićević et al ., 2020), whilst the second is based on the neural network technique (Malinovic‐Milicevic et al ., 2018).…”