Modeling direct irradiance from GOES visible channel using generalized cloud indices

Many simple operational models predicting global irradiance from the visible channel of geostationary satellites are based on the observation that the relationship between planetary albedo and atmospheric transmissivity is linear (Schmetz, 1989).. Current operational models (e.g., Zelenka et al., 1999) use the history normalized pixels at any given ground point to establish a dynamic range. The top of the range (brightest pixels) corresponds to heavily overcast conditions while the bottom of the range (darkest pixels) corresponds to clear conditions. Incoming pixels are gauged against this dynamic range, defining a cloud index. Global irradiance is linearly derived from this cloud index and quantified as a fraction of an appropriate clear sky model (e.g., Kasten, 1984).

Direct irradiance is generally derived as a byproduct of global using transposition models, many of which have been developed for solar energy applications (e.g., Perez et al., 1993). The compounding of models reduces achievable precision.

One-step pixel-to-direct irradiance: The above model requires a dynamic range that is independent of all solar geometry effects so that cloud indices derived from it only reflect current insolation conditions. The above model also assumes a linear relationship between the cloud index and the irradiance index. The difficulty of deriving a totally solar geometry-independent index led Ineichen and Perez (1999) to introduce a generalized cloud index and corresponding dynamic range that could vary with solar geometry and doesn't need to be linearly related to the considered irradiance index. The effectiveness of this generalized cloud index becomes apparent for components other than global irradiance, such as direct or diffuse, because the relationship of these components' indices against the pixel-derived cloud index is not linear. A one-step model can thus be conceived whereby the ratio of direct irradiance to clear sky direct irradiance is simply derived from the cloud index using a non-linear relationship varying with solar geometry.

The paper presents an initial validation of this one-step model against high quality hourly ground measurements in Albany, NY, covering a 9-month period. Results are compared to the traditional approach compounding two models to derive direct irradiance. The satellite data are hourly intermediate resolution visible images distributed by Unidata (1999).

The paper also discusses the implications of ground albedo variability and satellite calibration drifts on model performance. Finally the paper shows how readily, or potentially, available ancillary information (e.g., cloud cover from the National Weather Service) could lead to operational performance improvements.