Uncertainty Propagation in Ensemble Rainfall Prediction Systems used for Operational Real-Time Flood Forecasting

Abstract

Advances in mesoscale numerical weather prediction make it possible to provide quantitative precipitation forecasts (QPF) along with many other data fields at increasingly higher spatial resolutions. It is currently possible to incorporate high-resolution NWPs directly into flood forecasting systems in order to obtain an extended lead time. It is recognised, however, that direct application of rainfall outputs from the NWP model contributes considerable uncertainty to the final river flow forecasts as the uncertainties inherent in the NWP are propagated into hydrological and hydraulic domains and can also be magnified by the scaling process. As more and more “modern” flood forecasting systems are adopting this coupled approach, it is necessary to study uncertainty propagation and interaction between the NWP and the real-time flood forecast system model cascade, which terminates technically with the decision support system (DSS).

In this study, analysis is conducted to investigate the uncertainties in rainfall predictions that form the primary perturbation in a coupled NWP-hydrological model context. The ensemble method is used to account for both uncertainties due to incorrect/inaccurate initial conditions and those derived from model structure. Conventional statistics are employed to show variations over domains as well as point-wise targets. An adapted empirical orthogonal function analysis based upon principal components (EOF/PCA) is used to measure the diversity of ensemble members, which in turn provides a way to reconstruct a composite scenario that embodies most of the significant characteristics of ensemble forecast fields. The analyses of a typical ensemble QPF case over the catchment scale reveals that, although the NWP-based QPF can generally capture the rainfall pattern, uncertainties in rainfall at the scale of model grid relative to the catchment scale were always significant. Therefore, a cautious approach should be taken before the QPF, either deterministic or ensemble based, is injected into a flood forecasting system. Detailed results are discussed and comments made regarding the uncertainty propagation and the usability of the NWP-based QPF in the context of real-time flood forecasting systems.