Operational Hydrological Forecasting during the IPHEx-IOP Campaign – Meet the Challenge

An operational streamflow forecasting testbed was implemented during the Intense Observing Period (IOP) of the Integrated Precipitation and Hydrology Experiment (IPHEx-IOP) in May-June 2014 to characterize flood predictability skill in complex terrain and to investigate the propagation of uncertainty in quantitative precipitation forecasts (QPFs) and estimates (QPEs) to streamflow forecast uncertainty using a distributed hydrologic model. Specifically, hydrological forecasts for the 24 hour period beginning at 12:00 UTC on the current date were issued daily for 12 headwater catchments in the Southern Appalachians with drainage size ranging from 71km2 to 520 km2 using the Duke Coupled surface-groundwater Hydrology Model (DCHM) forced by 24-hour atmospheric conditions and hourly QPFs produced by the NASA-Unified Weather Research and Forecasting (NU-WRF) model. Radar-based QPE products were used to produce previous day hindcasts including initial conditions (e.g. soil moisture) for the present day forecasts. This manuscript first describes the operational testbed framework and workflow during the IPHEx-IOP including a synthesis of results, and second various data assimilation approaches are explored a posteriori (post-IOP) to improve operational (flash) flood forecasting. Although the IPHEx operational testbed results were promising in terms of not having missed any of the flash flood events during the IOP with large lead times of up to 6 hours, significant errors of overprediction or underprediction were identified that could be traced back to the QPFs and subgrid-scale variability of radar QPEs. To improve operational flood prediction, three data-merging strategies were pursued post-IOP: 1) to improve QPFs through assimilation of satellite-based microwave radiances into NU-WRF; 2) to improve QPEs by merging ground raingauge observations using simple but effective adjusting methods, and 3) to improve streamflow forecasts by assimilating river discharge observations into the DCHM using the Ensemble Kalman Filter (EnKF), the fixed-lag Ensemble Kalman Smoother (EnKS) and the Asyncronous EnKF (i.e. AEnKF) data assimilation techniques with noted success. Specifically, the NU-WRF produced QPFs with much better spatial rainfall patterns; hydrologic simulations forced by ensembles of merged QPEs from satellite and ground-based radar observations produce streamflow hindcasts and associated uncertainty that capture the observations well; and both the flood hindcast and forecast results were significantly improved by assimilating discharge observations into the DCHM. Hindcast Nash-Sutcliff Efficiency (NSE) values as high as 0.98, 0.71 and 0.99 at 15-min time-scales were attained for three headwater catchments in the inner mountain region demonstrating that the assimilation of discharge observations at the basin's outlet can reduce the errors and uncertainties in initial soil moisture. Success in operational flood forecasting at lead times of 6, 9, 12 and 15hrs was also achieved through discharge assimilation with NSEs of 0.87, 0.78, 0.72 and 0.51, respectively. These experiments indicate that the optimal assimilating time window depends both on basin properties and storm-specific space-time-structure of rainfall, and therefore adaptive, context-aware, configurations of the data assimilation system should prove useful to address the challenges of flood prediction in headwater basins.