Automatic state updating for operational streamflow forecasting via variational data assimilation

In operational hydrologic forecasting, to account for errors in the initial and boundary conditions, and in parameters and structures of the hydrologic models, the forecasters routinely make adjustments in real-time to the hydrometeorological input, hydrologic model states and, in certain cases, model parameters based on streamflow observations. Though a great deal of effort has been made in recent years to automate such “run-time modifications” (MOD) by human forecasters to a possible extent, automatic state updating of hydrologic models is yet to be widely accepted or routinely practiced in operational hydrology for a range of reasons. In this paper, we describe a state updating procedure intended specifically for operational streamflow forecasting for gauged headwater basins, and compare its performance with human forecaster MOD through a real-time forecasting experiment. The procedure is based on variational assimilation (VAR) of streamflow, precipitation and potential evaporation (PE) data into lumped soil moisture accounting and routing models operating at a 1-h timestep. The procedure has been in experimental operation since 2003 at the National Weather Service’s (NWS) West Gulf River Forecast Center (WGRFC) in Fort Worth, TX. Also described is a novel parameter estimation and optimization tool, the Adjoint-Based OPTimizer (AB_OPT), used for lumped hydrologic modeling at a 1-h timestep necessary for VAR.