The massive integration of stochastic renewable generation in modern power systems requires decentralized control schemes that cope with the uncertainties of the distributed energy resources (DERs). Among the DERs, small-scale photovoltaic (PV) systems are expected to represent most of the future power generation capacity. Therefore, solar resource assessment and the associated forecasting at various time scales are of fundamental importance for power systems operators. The activity here summarized focuses on developing forecasting methods based on the integrated use of time series, all-sky camera images and generation models of PV systems, considering short-term temporal horizons (below one hour) and fine spatial resolution (single site installations). We aim to improve the efficiency of forecasting methods developed by the DESL through prediction of clouds movement and associated cover of the sun disk. For this, we developed and validated different machine-learning techniques relying on the integrated use of Global Horizonal Irradiance (GHI) time series, all-sky image processing and cloud motion identification. Furthermore, a methodology to estimate the irradiance from all-sky images is proposed, investigating the possibility of using all-sky cameras as irradiance sensors.