From Point to Area: Worldwide Representativeness of Monthly Surface Solar Radiation Records

The spatial-temporal representativeness of surface solar radiation (SSR) point observations for a larger surrounding is a central issue when using these measurements in combination with gridded data, e.g. from satellite observations or numerical models. As previous studies on SSR representativeness focus mainly on individual aspects of representativeness and on small-scale observational networks, a global integrated assessment for the representativeness of SSR point observations is lacking. Consequently, while SSR point observations are broadly used in combination with gridded data as SSR is most accurately determined using ground-based observations, uncertainties arising from the imperfect spatial representativeness of SSR point data are often neglected which may substantially affect results. Through analyzing high resolution satellite-derived SSR data sets, we for the first time provide an integrated near-global (50°S to 55°N) assessment of the spatial and temporal representativeness of monthly SSR observations with focus on three different aspects of representativeness: decorrelation lengths (δ), spatial sampling biases (β), and spatial sampling errors (ε; defined on the 95%-level). Across the observational domain, we find a median δ larger than 3° which generally justifies the combination of point and gridded SSR data. For β and ε, which we analyze for a one-degree grid, we find a domain median of 1.4 and 7.6Wm^-2, respectively. These numbers are, however, not uniformly applicable as (a) regional differences are large and (b) they quantify typical (68.2%) sampling biases and errors for all locations within one-degree boxes. Location-specific values can be substantially larger or smaller. Therefore, we recommend using the location-specific errors and biases presented here. Comparing the domain median ε with the uncertainty from SSR measurement devices (i.e. 8Wm^-2) reveals that the sampling error (with respect to a one-degree grid) is of the same magnitude as the measurement uncertainty of state of the art SSR measurement devices. Thus, this study shows that the combination of point observations of monthly mean SSR time series with one-degree gridded data is justified within reasonable uncertainty in most regions, at the price of a combined uncertainty (sampling plus devices uncertainty) that is roughly 40-50% larger than the measurement devices' uncertainty alone.