Satellite image-based generation of high frequency solar radiation time series for the assessment of solar energy systems

Abstract

Solar energy is envisaged as a major pillar of the global transition to a climate-friendly energy system. Variability of solar radiation requires additional balancing measures to ensure a stable and secure energy supply. In order to analyze this issue in detail, solar radiation time series data of appropriate temporal and spatial resolution is necessary. Common weather models and satellites are only delivering solar surface irradiance with temporal resolutions of up to 15 min. Significant short-term variability in irradiances within seconds to minutes however is induced by clouds. Ground-based measurements typically used to capture this variability are costly and only sparsely available. Hence, a method to synthetically generate time series from currently available satellite imagery is of value for researchers, grid operators, and project developers. There are efforts to increase satellite resolution to 1 min, but this is not planned everywhere and will not change the spatial resolution. Therefore, the fundamental question remains if there are alternative strategies to obtain high temporal resolution observations at a pinpoint. This paper presents a method to generate 1 min resolved synthetic time series of global and direct normal irradiances for arbitrary locations. A neural network based on satellite image derived cloud structure parameters enables to classify high-frequency solar radiation variability. Combined with clear-sky radiation data, 1 min time series which reflect the typical variability characteristics of a site are reproduced. Testing and validation against ground observations (BSRN) show that the method can accurately reproduce characteristics such as frequency and ramp distributions. An application case demonstrates the usage in low-voltage grid studies.