Streamflow monitoring is very important for planning and management of water resources in watersheds, and their prediction accuracy is crucial for decision-making. The Niger River Basin is a transboundary resource, shared by nine West African Countries and Algeria and, a large portion of the population rely on the basin for rain-fed agriculture and hydropower. Over the years, the basin’s streamflow regime has been altered due to climate change, drought, desertification and establishment of Dams. This research describes a novel Deep Learning framework comprised of Bidirectional-Long Short-Term Memory (LSTM) requiring Antecedent Precipitation Index (API) and meteorological variables, preprocessed using Normal Quantile Transform (NQT) as input drivers and, compared with the Soil and Water Assessment Tool (SWAT+) for streamflow prediction. NQT-API-LSTM which considers catchment wetness and seasonality, was forced with reanalyzed climate (1979–2021) while, SWAT+ was driven with biophysical data and reanalyzed climate (2010–2020). The very high performance of both NQT-API-LSTM and SWAT+ models showed the models were reliable and can predict regulated flows with reasonable certainty. However, NQT-API-LSTM outperformed SWAT+ at Lokoja watershed and, realistically captured the influence of seasonal climate and regional groundwater dynamics from upstream catchments including the Sahara Desert on the Benue, Guinean, Sahelian and Sudan Flood. Overall, NQT-API-LSTM could be used successfully for watershed-scale streamflow prediction without the need for continuous ground support data, a benefit for sparsely gauged West African River Basins, while SWAT+ could be used as an alternative, particularly, to evaluate the watershed’s response to land use/land cover changes.