Variation among solar radiation data sets for the Eastern US and its effects on predictions of forest production and water yield

Spatially explicit modeling of ecosystem response to environmental change is often constrained as much by the accuracy of the data planes used to drive the model as by the differences in structure and parameterization of the models employed. In humid zones models of forest production at the site level are often more sensitive to solar radiation inputs than to other climate drivers. Radiation data are the most sparsely measured of routine climatological data. Still, we identified 3 different Web-based sources, containing 5 long-term mean radiation data sets, which differed in method of development and extrapolation. Comparisons are presented among the 5 data sets for estimated monthly mean total radiation as a fraction of total potential. The 5 vary by more than 20 % in mean monthly solar radiation and result in differences in predicted woody biomass production of as much as 21%, up to 12% in total net primary production. We selected 1 data set to develop a statistical model of mean monthly radiation in units relevant to photosynthesis models (photosynthetically active radiation [PAR], pmol m(-2) s(-1)) for the Eastern US, as part of the USDA Forest Service Eastwide global change effort (a joint effort of the Northern and Southern Global Change Programs). A simple multiple linear model predicted spatial PAR values with a standard error of the estimate of less than 6 % in all months, and with lowest values (<3%) in mid-year.