Reliable seasonal precipitation forecasts are vital for managing hydroelectric power plants, particularly in regions with variable climate conditions. This study evaluates the performance of North American Multi-Model Ensemble (NMME) wet-season precipitation forecasts over two hydroelectric basins in Brazil: Itaipu, characterised by a humid subtropical climate, and Sobradinho, located in a semi-arid region. By assessing six NMME models against Integrated Multi-satellitE Retrievals for GPM (IMERG) data and comparing them with statistical models based on atmospheric-oceanic indices, the study identifies significant spatial and model-dependent variations in forecast skill. NMME models struggle with regional anomalies and extreme events, exhibiting systematic biases and limited predictive capability, particularly in drought-prone Sobradinho. In contrast, statistical models leveraging El Ni & ntilde;o-Southern Oscillation (ENSO), North Atlantic Oscillation (NAO), Tropical Northern Atlantic Index (TNA), and Tropical Southern Atlantic Index (TSA) indices demonstrate better predictive accuracy. Incorporating select NMME models as predictors improves statistical model performance, highlighting the potential of hybrid modelling approaches. The results emphasise the need for improved parameterisations, localised data integration, and machine learning-driven enhancements to refine seasonal precipitation forecasts for hydropower-critical regions.