Abstract
Daily insulation values at the earth's surface are required for modeling of biophysical processes and solar energy engineering design. Ground-based pyranometer networks have proliferated in recent years, offering improved spatial coverage for some regions, although data assimilation is daunting. Geosynchronous satellites can provide high-spatial-resolution observations, although data are not routinely available and estimates for snow-covered areas are problematical. The adequacy of the present array of ground-based pyranometers in the Midwest for estimating daily insulation was tested using kriging, a geostatistical technique. First, the usefulness of the kriging variance for predicting likely errors in spatial interpolation was demonstrated using a high-spatial-resolution satellite-derived insulation dataset. Then, daily data for 51 locations in four states were screened for errors using double-mass plots, a technique widely used in hydrology. For each day, a variogram was fit to the observations (usually termed a structure function in atmospheric sciences) and used to estimate values on a regular grid, using kriging. Likely uncertainty on kriged estimates of insulation increased with distance from the unknown point to the nearest measurement site. Uncertainty was also a function of areally averaged insulation: clear or overcast days had smaller errors than did days of intermediate insulation, presumably because the intermediate days had mixed clear and cloudy areas within the region. For sites 100 km from the nearest measurement, clear days had uncertainties between 1.5 and 3.5 MJ m−2 day−1, cloudy days ranged from 0.5 to 3.5, and intermediate days had errors from 0.5 to 4.4 MJ m−2 day−1.
