ABSTRACT

In this talk, the Pennsylvania State Regional Hydrologic Ensemble Prediction System (RHEPS) is demonstrated. The RHEPS is a research forecasting system aimed at enhancing hydrologic forecasting at regional spatial scales (~10²-10⁴ km²) by facilitating the comprehensive and integrated evaluation of different system components. The aim is also for the RHEPS to support the research-to-operation transition by mimicking and improving specific aspects of operational forecasting systems, and training the next generation of hydrologic forecasters. The RHEPS is, at present, comprised by the following system components: ensemble meteorological forcing (i.e. output fields from numerical weather prediction models), statistical weather postprocessor, multimodel hydrologic-hydraulic framework, statistical streamflow postprocessor, and verification framework. The application of the RHEPS and some of its components will be demonstrated in this talk using two case studies. The two case studies are focused on the U.S. middle Atlantic region (MAR) and rely on meteorological forcing (precipitation and land surface temperature outputs) from a single model and single physics ensembles. Specifically, outputs from the National Centers for Environmental Prediction 11-member Global Ensemble Forecast System Reforecast version 2 (GEFSRv2) are used since the GEFSRv2 provides a long reforecast dataset useful for verification purposes.

The first case study demonstrates the implementation of the RHEPS for streamflow forecasting over some of the major river basins in the MAR, including the Delaware, James, Potomac, and North Brach Susquehanna River, for forecast lead times of 6-168 h. For this case study, together with the GEFSRv2 outputs, the NOAA’s Hydrology Laboratory Research Distributed Hydrologic Model (HL-RDHM) is used as the distributed hydrologic model and an auto-regressive time series model as the statistical streamflow postprocessor. Results from this case study highlight that the ensemble streamflow forecasts are more skillful than deterministic ones and remain skillful up to lead times of 7 days, and that postprocessing enhances forecast skill.

The second case study is focused on the statistical weather postprocessor component of the RHEPS. Using precipitation outputs from the GEFSRv2, Bayesian model averaging (BMA) and heteroscedastic censored logistic regression (HCLR) are assessed over the MAR, using different metrics (e.g., skills scores and reliability diagrams) conditioned upon the forecast lead time, precipitation threshold, and season. Overall, HCLR tends to slightly outperform BMA but the differences among the postprocessors are not as significant. A shortcoming of HCLR, however, is that it does not provide a means for combining (weighting) ensemble forecasts from different models. In the future, an alternative approach could be to combine HCLR with BMA to take advantage of their relative strengths. The talk will conclude with some additional recommendations to continue to advance the RHEPS and hydrologic forecasting in general over the MAR.