<p>Solar energy proves to be an interesting alternative to conventional sources based on the burning of fossil fuels. However, it shows a high short-term variability that makes its integration into the electricity mix difficult. To facilitate this integration, reliable short- and medium-term forecasts become highly necessary. To respond to this demand, solar radiation forecasting models have emerged. Among them, Weather Research and Forecasting (WRF) has become particularly promising and has shown good performance at different temporal and spatial scales. The performance of these models is usually assessed by comparing their estimates with point measurements at selected stations. This comparison is hampered by the difference in spatial dimensions between the model estimates (representative of a given area) and the station (point) measurements. This difference introduces a certain error in the forecast, mainly related to the short-scale variability of cloudiness. Despite being essential to understand model validation, this issue has not been sufficiently investigated.&#160;In this framework, the present study analyzes the effect of the spatial representativeness of point measurements when used to validate model estimates. For this purpose, a specific one-month measurement campaign was conducted, deploying seven pyranometers in the vicinity of the city of Badajoz, Spain. To ensure their intercomparability, all pyranometers were calibrated with respect to a reference pyranometer previously calibrated by the World Radiation Center in Davos, Switzerland. Solar radiation was measured at a 1-minute basis to record the short-term variability due to cloudiness. Two series were constructed with these data, one corresponding to a selected station and the other to the average of the seven stations.&#160;These series of measurements were compared with the estimates provided by the WRF model for the same period and location. A configuration with two nested domains of 27 km and 9 km was used. Model performance showed better agreement when averaging was used instead of individual measurements, with RMSE improving from 89 W/m&#178; to 77 W/m&#178;. Cloudy cases contributed the most to the differences between station measurements and model estimates, showing an RMSE greater than 100 W/m2, more than three times higher than the RMSE for clear cases (about 33 W/m2). The difference between the stations and the model for cloudy cases is reduced from 125 W/m2 to 107 W/m2 when averaged measurements are considered instead of single station measurements. This study contributes to the understanding of the representativeness of point station measurements for validation and comparison with WRF estimates.&#160;Acknowledgments. This work is partially funded by FEDER/Ministerio de Ciencia, Innovaci&#243;n y Universidades-Agencia Estatal de Investigaci&#243;n of Spain through project RTI 2018-097332-B-C22, and by Junta de Extremadura-FEDER through project GR18097.</p>
An Object-Oriented Multiscale Verification Scheme
