An extension of data assimilation into the short-term hydrologic forecast for improved prediction reliability

Typically, when using data assimilation to improve hydrologic forecasting, observations are assimilated up to the start of the forecast. This is done to provide more accurate state and parameter estimates which, in turn, allows for a better forecast. We propose an extension to the traditional data assimilation approach which allows for assimilation to continue into the forecast to further improve the forecast's performance and reliability. This method was tested on two small, highly urbanized basins in southern Ontario, Canada; the Don River and Black Creek basins. Using a database of forcing data, model states, predicted streamflow, and streamflow observations, a lookup function was used to provide an observation during the forecast which can be assimilated. This allows for an indirect way to assimilate the numerical weather prediction forcing data. This approach can help in addressing prediction uncertainty, since an ensemble of previous observations can be pulled from the database which correspond to the forecast probability density function given previous information. The results show that extending data assimilation into the forecast can improve forecast performance in these urban basins, and it was shown that the forecast reliability could be improved by up to 78%.