The article introduces novel ensemble data assimilation (DA) techniques for streamflow forecasting using WRF-Hydro and DART. Model-related biases are tackled through spatially and temporally varying adaptive prior and posterior inflation. Spurious and physically incorrect correlations, on the other hand, are mitigated using a topologically based along-the-stream localization. Hurricane Florence (2018) in the Carolinas, USA, is used as a test case to investigate the performance of DA techniques.