Iterative filtering of ground data for qualifying statistical models for solar irradiance estimation from satellite data Jesus Polo * , Luis F. Zarzalejo, Lourdes Ramirez, Bella Espinar Renewable Energy Department, CIEMAT, Avad. Complutense, 22, 28040 Madrid, Spain Received 26 January 2004; received in revised form 17 February 2005; accepted 9 March 2005 Available online 11 April 2005 Communicated by: Associate Editor Pierre Ineichen Abstract A new technique of filtering solar radiation ground data is proposed for generating models for solar irradian mation from geostationary satellite data. The filtering processes consists of an iterative way of selecting the training data set to achieve the best model response. Although in this paper the proposed methodology has been used irradiance modeling, it could be applied to any kind of empirical modeling. The iterative filtering method has p have fast convergence and to improve successfully the statistical model response, when applied to hourly glo diance calculation from satellite-derived irradiances for 13 Spanish locations. Individual statistical models for h global irradiance were fitted using the Heliosat I method applied to Meteosat images of 13 Spanish stations fo iod 1994–1996. Ó 2005 Elsevier Ltd. All rights reserved. Keywords: Solar irradiance; Meteosat satellite; Active learning; Ground database quality 1. Introduction The quantification ofthe solar irradianceat the earthÕs surface is of great interest in solar energy, mete- orology and climatic applications. However, the scarcity of ground data for many regions has required the use of geostationary satellite data as an accurate tool for esti- mating solarradiation with a spatial resolution ofa few kilometers. Several authors have developed methods for estimating the cloud index from satellite data ( Tarp- ley, 1979; Cano et al., 1986; Diabate et al., 1988). The different models available for estimating the so- lar irradiance from satellite images can be grouped into two different approaches, statistical and physical (Noia et al., 1993). Statistical methods are based mainly on sta tistical regressions between satellite count and the corre- sponding measurement at the earthÕs surface. Physical methodsare based exclusively on physical consider- ations of how the radiantenergy interacts with the earthÕs atmosphere. Both approacheshave particular advantages and shortcomings, and thus the choice of which approach to use in the development of a model depend,to a greatextent,on the data available. On the one hand,the physicalmodelapproach requires complementary meteorological data to estimate several parameters related to the interaction of solar radiation with the atmosphere. On the other hand, the statistical 0038-092X/$ - see front matter Ó 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.solener.2005.03.004 * Corresponding author. Tel.: +34 91 3466043; fax: +34 91 3466037. E-mail address: jesus.polo@ciemat.es (J. Polo). Solar Energy 80 (2006) 240–247 www.elsevier.com/locate/solener