Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Hydrology and Remote Sensing Laboratory » Research » Publications at this Location » Publication #300758
Title: An integrated error estimation and lag-aware data assimilation scheme for real-time flood forecasting

Author
item LI, YUAN - University Of Melbourne
item RYU, D - University Of Melbourne
item WESTERN, A - University Of Melbourne
item WANG, Q - Collaborator
item ROBERTSON, D - Collaborator
item Crow, Wade

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/10/2014
Publication Date: 11/1/2014
Publication URL: http://handle.nal.usda.gov/10113/60333
Citation: Li, Y., Ryu, D., Western, A., Wang, Q.L., Robertson, D., Crow, W.T. 2014. An integrated error estimation and lag-aware data assimilation scheme for real-time flood forecasting. Journal of Hydrology. 519(D):2722-2736. DOI: 10.1016/j.jhydrol.2014.08.009.

Interpretive Summary: Stream flow forecasts are valuable for a range of agricultural management applications including: flood forecasting, irrigation scheduling and efforts to minimize the water quality impact of fertilizer application. Currently, the most accurate way to acquire such forecasts is via the use of hydrologic data assimilation systems which optimally combine model predictions of stream flow with real-time observations of rainfall and stream flow stage. However, the performance of these data assimilation approaches is very sensitive to our ability to statistically describe errors in both hydrologic models and hydrologic observations. A better statistical representation of these errors will inevitably lead to more accurate data assimilation predictions. This paper describes a new mathematical approach for estimating hydrologic error statistics. Results suggest that the operational implementation of the approach can potentially improve the quality of stream flow forecasts available in agricultural watersheds.

Technical Abstract: The performance of conventional filtering methods can be degraded by ignoring the time lag between soil moisture and discharge response when discharge observations are assimilated into streamflow modelling. This has led to the ongoing development of more optimal ways to implement sequential data assimilation methods for operational flood forecasting. In this paper, an ensemble Kalman smoother (EnKS) with fixed time window is implemented for the GR4H model (modèle du Génie Rural à 4 paramètres Horaire) to update current and antecedent states, and is compared with the standard ensemble Kalman filter (EnKF). The model and observation error parameters are estimated through the maximum a posteriori (MAP) calibration approach constrained by prior information drawn from flow gauging data. The updating schemes assimilate real discharge observations and they are evaluated in both synthetic forecasting mode with observed rainfall and real-time forecasting mode with the Numerical Weather Prediction (NWP) forecast rainfall. The results show that the EnKS outperforms the EnKF in synthetic forecasting mode by providing more accurate discharge forecasts. The EnKS is more stable then the EnKF, which tends to overcorrect the current states. In real-time forecasting mode, the benefit from state updating is reduced because the uncertainty in the NWP rainfall data becomes dominant with increasing forecasting lead time.