Articles | Volume 17, issue 11
https://doi.org/10.5194/acp-17-7157-2017
https://doi.org/10.5194/acp-17-7157-2017
Research article
 | 
16 Jun 2017
Research article |  | 16 Jun 2017

TEMIS UV product validation using NILU-UV ground-based measurements in Thessaloniki, Greece

Melina-Maria Zempila, Jos H. G. M. van Geffen, Michael Taylor, Ilias Fountoulakis, Maria-Elissavet Koukouli, Michiel van Weele, Ronald J. van der A, Alkiviadis Bais, Charikleia Meleti, and Dimitrios Balis

Abstract. This study aims to cross-validate ground-based and satellite-based models of three photobiological UV effective dose products: the Commission Internationale de l'Éclairage (CIE) erythemal UV, the production of vitamin D in the skin, and DNA damage, using high-temporal-resolution surface-based measurements of solar UV spectral irradiances from a synergy of instruments and models. The satellite-based Tropospheric Emission Monitoring Internet Service (TEMIS; version 1.4) UV daily dose data products were evaluated over the period 2009 to 2014 with ground-based data from a Norsk Institutt for Luftforskning (NILU)-UV multifilter radiometer located at the northern midlatitude super-site of the Laboratory of Atmospheric Physics, Aristotle University of Thessaloniki (LAP/AUTh), in Greece. For the NILU-UV effective dose rates retrieval algorithm, a neural network (NN) was trained to learn the nonlinear functional relation between NILU-UV irradiances and collocated Brewer-based photobiological effective dose products. Then the algorithm was subjected to sensitivity analysis and validation. The correlation of the NN estimates with target outputs was high (r = 0. 988 to 0.990) and with a very low bias (0.000 to 0.011 in absolute units) proving the robustness of the NN algorithm. For further evaluation of the NILU NN-derived products, retrievals of the vitamin D and DNA-damage effective doses from a collocated Yankee Environmental Systems (YES) UVB-1 pyranometer were used. For cloud-free days, differences in the derived UV doses are better than 2 % for all UV dose products, revealing the reference quality of the ground-based UV doses at Thessaloniki from the NILU-UV NN retrievals. The TEMIS UV doses used in this study are derived from ozone measurements by the SCIAMACHY/Envisat and GOME2/MetOp-A satellite instruments, over the European domain in combination with SEVIRI/Meteosat-based diurnal cycle of the cloud cover fraction per 0. 5° × 0. 5° (lat  ×  long) grid cells. TEMIS UV doses were found to be ∼ 12.5 % higher than the NILU NN estimates but, despite the presence of a visually apparent seasonal pattern, the R2 values were found to be robustly high and equal to 0.92–0.93 for 1588 all-sky coincidences. These results significantly improve when limiting the dataset to cloud-free days with differences of 0.57 % for the erythemal doses, 1.22 % for the vitamin D doses, and 1.18 % for the DNA-damage doses, with standard deviations of the order of 11–13 %. The improvement of the comparative statistics under cloud-free cases further testifies to the importance of the appropriate consideration of the contribution of clouds in the UV radiation reaching the Earth's surface. For the urban area of Thessaloniki, with highly variable aerosol, the weakness of the implicit aerosol information introduced to the TEMIS UV dose algorithm was revealed by comparison of the datasets to aerosol optical depths at 340 nm as reported by a collocated CIMEL sun photometer, operating in Thessaloniki at LAP/AUTh as part of the NASA Aerosol Robotic Network.

Download
Short summary
NILU irradiances at five UV channels were used to produce CIE, vitamin D, and DNA- damage daily doses via a neural network (NN) model. The NN was trained with collocated weighted Brewer spectra and uncertainty in the NILU-derived UV effective doses was 7.5 %. TEMIS UV products were found to be ~ 12.5 % higher than the NILU estimates. The results improve for cloud-free days with differences of 0.57 % for CIE, 1.22 % for vitamin D, and 1.18 % for DNA damage, with standard deviations of ~ 11–13 %.
Altmetrics
Final-revised paper
Preprint