Accurate solar resource information is a fundamental requirement for solar energy ventures. The lack of precision in solar radiation data can significantly affect the success of the projects. Argentina has solar radiation ground measurement networks. The information obtained through this method is limited due to its spatial sparsity, since it is only possible to measure with appropriate quality in some sites across the territory. To overcome this limitation, it is common to generate models capable of estimating solar radiation through satellite images, which provide spatial resolution. This work develops and validates an empirical model for this purpose based on Machine Learning (ML), demonstrating that it is a useful and accurate tool to be considered. This allows ventures that make use of this type of energy to have greater certainty in the availability of the resource, and therefore in the decision-making process. Variables obtained from images of the geostationary meteorological satellite GOES-16, McClear clear-sky model estimates, and geometrically calculated information are used as input to the algorithms. The results of the ML models are compared with estimates from pre-existing models for the region that incorporate physical modelings, such as Heliosat-4 and CIM-ESRA. The evaluation shows a higher performance of the ML methods when multi-scale satellite information is used as input. The incorporation of multi-scale satellite data is not yet implemented in solar radiation physical modeling, which is an advantage of ML modeling.