Epidemiological mathematical models have been proved crucial in supporting the decision-making of the health authorities during the COVID-19 pandemic. In this context, this work presents two contributions. The first one is a methodology to integrate different data sources into a single time series that provides realistic COVID-19 incidence rates considering both the reported and unreported cases in Spain and Comunidad de Madrid. The second contribution is a novel ensemble forecast model that uses as input the predictions of three different COVID-19 forecasts models. These approaches have been used to provide forecast predictions in the scope of PredCov project, supporting both the Spanish and the European Union -via the European Centre for Disease Prevention and Control-health authorities. The output generated by the ensemble model provides a combined -and more accurate-prediction of the COVID-19 incidence. This work includes a description of both contributions and discusses the results provided by them.