Validation of NWS Hydrologic Ensemble Forecast Service (HEFS) Real-Time Products at the Middle Atlantic River Forecast Center

The National Weather Service (NWS) hydrology program has a long history of generating ensemble river forecasts for a range of applications. In the 1980s NWS River Forecast Centers began running the Extended Streamflow Prediction System (ESP). The ESP uses historical observations of precipitation and temperature to force hydrologic models and produce an ensemble of plausible water forecasts. ESP forecasts are typically 30 - 90 day forecasts and are geared towards water resources managers. In 2010, four NWS RFCs began producing 3 - 7 day ensemble forecasts using the Meteorological Model Ensemble Forecast System (MMEFS). MMEFS uses forecasts of precipitation and temperature from global meteorological models to force hydrologic models and produce ensemble forecasts. MMEFS has proven useful as a situational awareness tool for river flood potential.

In 2014, the Office of Water Prediction (OWP) released the Hydrologic Ensemble Forecast Service (HEFS) v.1.0 to further improve RFC ensemble forecasting capabilities. The HEFS provides modules to account for the uncertainty in model forcings and model structure. The HEFS forcings module (MEFP, Meteorological Ensemble Forecast Processor) can account for the uncertainty from multiple forcing sources, such as three-day deterministic quantitative precipitation forecasts, 15 day Global Ensemble Forecast System (GEFS) forecasts, and 270 day Climate Forecast System (CFSv2) forecasts. The MEFP attempts to correct for biases in the meteorological model forcings and also blends together forecasts over different lead times. The NWS Middle Atlantic River Forecast Center (MARFC) began issuing HEFS v.1.0 products for several watersheds of interest to New York City water supply managers in 2014. In 2017, MARFC began issuing 10 day probabilistic forecasts from HEFS at 115 locations. These 10 day products are of interest to both emergency managers (for stage forecasts) as well as water managers and hydropower operators in the region (for volume estimates).

OWP completed extensive validation of HEFS prior to implementation using selected basins across the country, including two in the MARFC region. This validation involved running hindcasts and analyzing results over a range of lead times from days to months. MARFC worked with OWP to run additional hindcasts and validation on all of our forecast points prior to implementing the 10 day products in 2017. The MARFC hindcasting studies provided a mechanism to check our software configuration, screen out bad historical data, and show the potential value of HEFS. However, this hindcast-based validation has two key limitations. First, these hindcasts do not account for initial condition updates that are made by forecasters in real-time. Second, HEFS skill was measured using an ESP-based approach as the reference forecast. For us to provide proper guidance to partners who have routinely used MMEFS products for several years, we need to provide information about the skill of HEFS relative to MMEFS.

In this paper, we report on validation of real-time HEFS forecast products from Jan 2017 through July of 2019. The analysis includes stage forecast validation at 103 gaged locations. To examine the HEFS and MMEFS forecast skill at high flows, we compute contingency statistics related to exceeding caution or flood staqe. We also examine statistical performance for low, medium, and high flows at a smaller number of locations of interest to water supply managers. The two main findings are as follows: (1) In terms of the critical success index (CSI) for high flows, HEFS does better in cold season but MMEFS does better in the warm season, (2) HEFS performs better for low to moderate flow ranges. A possible reason MMEFS performs relatively well in the warm season is because the system is more ‘bullish’ on flooding overall compared to HEFS, and both systems have a difficult time predicting warm-season floods driven by convective rains. Work to improve HEFS is ongoing, including an update to use GEFS v12, which was not used in these validated real-time products. Results here suggest that while our HEFS implementation continues to improve, there is still value in running MMEFS in parallel.