Spatio-temporal solar forecasting uses spatially distributed solar radiation or photovoltaic power data to enhance the forecasting at a given site. Two data sets with a wide range of time and spatial resolutions are explored using linear Auto-Regressive models with eXogenous inputs (ARX). Results allow the identification of two different forecasting modes of operation. A short-term mode, where suitable neighbours may significantly improve the forecasting performance, with skill values up to 30-40%, as they provide information on incoming clouds, and a longer-term mode, where the neighbouring sensors' positioning is less relevant as the positive skill values around 10-20% are associated to a spatial smoothing effect which reduces the occurrence of high forecast errors. For the short-term mode, the correlation between forecast horizons and effective distance to the most contributing neighbours was shown by a normalized weighted average distance (nWAD) parameter. Additionally, this parameter further sustained that the sensor network layout is not relevant for the second mode.