UV Reconstruction Techniques Anders Lindfors Finnish Meteorological Institute, Helsinki, Finland Abstract. UV measurement time series are, in general, relatively short. In order to extend our knowledge of UV radiation intensities to locations and times not covered by measurements, UV reconstruction techniques are needed. This paper presents an overview of currently available techniques for UV reconstruction. Introduction The scientific interest in surface UV radiation has a long history. Already in the early 20 th century, Dorno (1911) performed measurements of surface UV radiation at Davos. Although some measurements were performed also later on in the 20 th century, for instance, by Bener in the 1960s, recent UV measurements at Davos were not initiated until 1995. In order to cover the period in between, Lindfors and Vuilleumier (2005) reconstructed daily erythemally weighted UV doses at Davos over the period 1926-2003. Since UV measurement time series typically cover only the last 15 years, there is a need for reconstructed UV radiation also at other locations than Davos. Several methods for reconstruction of past UV radiation have been presented in the literature (see Calbo et al., 2005; Weatherhead et al, 2005; and references therein). This paper presents an overview of currently available techniques for UV reconstruction. Methods Long-term measurements of parameters that can be utilized for reconstruction of past UV radiation are available. The challenge from a UV reconstruction point of view is to find the best way to use this limited information. Most UV reconstruction techniques take as input at least the total ozone column and some data that can be used for describing the effect of clouds. Also the variations in the surface albedo may be accounted for, based on, for instance, measurements of snow depth. Because of lack of long-term measurements that could be used, aerosols are, in general, assumed constant or set to follow a predefined annual cycle. Statistical methods for reconstructing past UV radiation, where the complex relationship between the surface UV irradiance and the parameters given as input to the method are determined using statistical analysis, have been presented by, for instance, Fioletov et al. (2001), Feister et al. (2002), de la Casiniere et al. (2002), Diaz et al. (2003), and Krzyscin et al. (2004). Junk et al. (2007), in turn, developed their UV reconstruction method based on neural networks. Many techniques for reconstructing past UV radiation are based on first predicting the clear-sky UV using, for instance, radiative transfer calculations, and then correcting this value for the effect of clouds using a so- called cloud modification factor: UV reco = UV clear CMF UV (1) where UV reco is the reconstructed UV radiation, UV clear is its predicted clear-sky value, and CMF UV is the cloud modification factor for UV radiation (see, e.g., Lindfors et al., 2007 for a definition of CMF UV ). This equation highlights an interesting area when comparing different UV reconstruction techniques, namely the effect of clouds and how they are treated in each method. Lindfors et al. (2003) and Eerme et al. (2002), for instance, developed UV reconstruction methods based on empirical relationships found between CMF UV and the relative daily sunshine duration. Many others, for instance, Bodeker and McKenzie (1996), Kaurola et al. (2000), Krzyscin et al. (2004), Trepte and Winkler (2004), den Outer et al. (2005), Kazantzidis et al. (2006), and Lindfors et al. (2007) have used pyranometer measurements of global radiation (300- 3000 nm) for estimating CMF UV . Pyranometer data contain valuable information about the influence of clouds on radiation. Indeed, the most accurate UV reconstruction methods in a study by Koepke et al. (2006) were based on pyranometer measurements. Some methods are based on observations of the total cloud amount. For instance, Reuder and Koepke (2005) determined CMF UV using neural networks with total cloud amount as input. Engelsen et al. (2004) used a different approach. Instead of determining CMF UV they set the clouds in their radiative transfer calculations to resemble the typical optical properties of the observed clouds at the location of interest, thus allowing them to estimate the surface UV radiation based on cloud amount. Results and Discussion Many of the papers mentioned above have presented long time series of reconstructed UV radiation. Such time series are crucial for our understanding of the past UV radiation climate, and also put the recent changes driven by the ozone decline observed over the last few decades into perspective. Furthermore, such time series are expected to raise interest also among UV impact scientists. As demonstrated by Chubarova and Nezval (2000) and Lindfors and Vuilleumier (2005), UV reconstruction techniques allow attributing the changes in the UV radiation level to changes in other climate parameters such as total ozone, clouds or surface albedo. The length of the reconstructed time series is dependent on the available input data. Sunshine duration measurements and observations of the total cloud amount are both available from the early 20 th century, and even earlier at some locations. The longest time series of reconstructed UV, extending back to 1926 and 1936, respectively, have been presented by Lindfors and