Probabilistic Verification of the National Water Model

Ensemble streamflow forecasting provides a means of characterizing prediction uncertainty. Though the short-range forecasts from the National Water Model are deterministic, information across overlapping forecast cycles can be interpreted probabilistically, as an ensemble. This so-called “time-lagged ensemble” is experimental, but is already being interpreted qualitatively by forecasters. Here we aim to quantify the skill of the experimental time-lagged ensemble forecasts to further support forecasters and the development of forecast guidance. We evaluate National Water Model short-term (18 hour) streamflow forecasts using both probabilistic and deterministic verification metrics. Results are presented for the 2017 Flash Flood and Intense Rainfall Experiment (FFaIR), highlighting how probabilistic forecasts can be used towards the prediction of high-impact events. We point out how probabilistic metrics are distinctive from, as well as complementary to traditional deterministic metrics, highlighting some limitations as well as opportunities towards improving predictions.