Forecasting solar irradiance is a key component in the management of electrical grids with a high percentage of photovoltaic energy generation. The main challenge in solar irradiance forecasting comes from the stochastic nature of cloud advection and diffusion. This spatio-temporal phenomena requires spatio-temporal data to predict, a source of which is sky imagers; wide-angle lensed cameras facing upwards taking images of the sky at regular intervals. This study investigated the use of machine learning models trained on large historical datasets of sky images as a source of solar irradiance forecasts.