Due to their uncertain and variable nature, the large-scale integration of wind and solar power poses significant challenges to the generator scheduling process in power systems. To support this process, system operators require using repeatedly updated forecasts of the best possible quality for renewable power. Motivated by this, the present work aims to study the benefits of incorporating spatiotemporal dependence and seasonalities into probabilistic forecasts for the intraday unit commitment (UC) process. With this purpose, a highly efficient forecasting-optimization scheme is proposed, which is composed of a detrended periodic vector autoregressive model and a technology-clustered interval UC model. The proposed approach is tested on a 120-GW power system with 210 conventional generators using real wind and solar measurements and compared to existing deterministic and stochastic UC techniques alongside standard forecasting methods. Extensive computational experiments show that the incorporation of spatiotemporal dependence and seasonalities into forecasts translates in a reduction of up to 1.55% in operational costs for a daily UC relative to standard practice, the application of intraday instead of daily UC runs further reduces operational costs in up to 1.51%, and the proposed forecasting-optimization scheme takes less than 10 h to simulate a whole year.