Representativeness errors of point-scale ground-based solar radiation measurements in the validation of remote sensing products

We usually use ground-based solar radiation measurements to validate satellite-derived solar radiation products from kilometer to grid scales. Questions such as, how large is the representativeness error of surface measurements in the validation and how much of the product-measurement difference can be attributed to their inherent differing spatial scales, cast doubts on the suitability of this direct validation approach. In this paper, we will investigate and quantify the representativeness errors of point-scale ground-based measurements using the surface flux observation matrix from HiWATER (Li et al., 2013) and the solar radiation data retrieved from geostationary meteorological satellite (Huang, Li, Ma, & Li, 2016). The current study demonstrates that wildly fluctuating representativeness errors exist which are strongly contingent on the time and space scales of remote sensing products, as well as instant atmospheric conditions. For example, for an area of 5 x 5 km(2) 1.4-8.1% of representativeness errors are found from monthly to instantaneous timescales; while for an area of 1 degrees x 1 degrees grid 3.1-8.1% of representativeness errors are seen. Such scale-dependent representativeness errors offer some implications for validations of remote sensing products. On timescales longer than or equal to one day, representativeness errors do not need to be considered for validations of kilometer-level products, but on shorter timescales representativeness errors will affect the validation results to some extent. For instantaneous products with 5 km resolution, our study indicates over 13% of errors can be attributed to the inherent representativeness error, and 30-minute surface measurements are recommended for a routine validation. However, for validations of grid-level products, representativeness errors basically cannot be neglected regardless of timescales. The errors caused by the poor representativeness of surface sites, likely significantly contribute to the large differences between measurements and products. (C) 2016 Elsevier Inc. All rights reserved.