Efficient treatment of climate data uncertainty in ensemble Kalman filter (EnKF) based on an existing historical climate ensemble dataset

Abstract

Successful data assimilation depends on the accurate estimation of forcing data uncertainty. Forcing data uncertainty is typically estimated based on statistical error models. In practice, the hyper-parameters of statistical error models are often estimated by a trial-and-error tuning process, requiring significant analyst and computational time. To improve the efficiency of forcing data uncertainty estimation, this study proposes the direct use of existing ensemble climate products to represent climate data uncertainty in the ensemble Kalman filter (EnKF) of flow forecasting. Specifically, the Newman et al. (2015) dataset (N15 for short), covering the contiguous United States, northern Mexico, and southern Canada, is used here to generate the precipitation and temperature ensemble in the EnKF application. This study for the first time compares the N15 generated climate ensemble with the carefully tuned hyper-parameters generated climate ensemble in a real flow forecasting framework. The forecast performance comparison of 20 Québec catchments shows that the N15 generated climate ensemble yields improved or similar deterministic and probabilistic flow forecasts relative to the carefully tuned hyper-parameters generated climate ensemble. Improvements are most evident for short lead times (i.e., 1–3 days) when the influence of data assimilation dominates. However, the analysis and computational time required to use N15 is much less compared to the typical trial-and-error hyper-parameter tuning process.