Understanding the spatial and temporal characteristics of solar resource variability is important because it helps inform the discussion surrounding the merits of geographic dispersion and subsequent electrical interconnection of photovoltaics as part of a portfolio of future solutions for coping with this variability. Unpredictable resource variability arising from the stochastic nature of meteorological phenomena (from the passage of clouds to the movement of weather systems) is of most concern for achieving high PV penetration because unlike the passage of seasons or the shift from day to night, the uncertainty makes planning a challenge. A suitable proxy for unpredictable solar resource variability at any given location is the series of variations in the clearness index from one time period to the next because the clearness index is largely independent of the predictable influence of solar geometry. At timescales shorter than one day, the correlation between these variations in clearness index at pairs of distinct geographic locations decreases with spatial extent and with timescale. As the aggregate variability across N decorrelated locations decreases as 1/root N, identifying the distance required to achieve this decorrelation is critical to quantifying the expected reduction in variability from geographic dispersion. Using 10 years of satellite-derived daily-interval solar resource data across the world, we demonstrate that the spatiotemporal behavior of unpredictable solar resource variability is mirrored at longer tirnescales. We do so by examining over 1.4 million unique pairs of sites across the Western hemisphere and quantifying the influence each pair's geographic separation and bearing has on the correlation between the variability of each pair's clearness indices at timescales of one, two, four, seven, fifteen and thirty days. Expected pair-decorrelation distances are estimated by fitting exponential trends to the data using nonlinear least-squares regression and are presented as a function of timescale and pair orientation. Reflecting the predominant direction in which meteorological phenomena propagate at each of these timescales, we find that pairs of sites require considerably shorter distances to decorrelate when they are oriented north to south versus when they are oriented east to west. As at shorter timescales, these decorrelation distances are shown to increase with both timescale and with geographic extent. (C) 2015 Elsevier Ltd. All rights reserved.