Monday, 7 January 2019: 12:00 AM
North 126BC (Phoenix Convention Center - West and North Buildings)
Streamflow predictions include errors from both the magnitude and timing of hydrologic events. Nevertheless, common model evaluation metrics, such as root mean square error (RMSE) and Nash-Sutcliffe efficiency (NSE), don’t necessarily distinguish between different types of errors nor errors occurring at different time and/or spatial scales. Timing error estimates are rarely explicitly provided, despite the potential utility of this information for forecast guidance and model development.
Wavelet-based approaches decompose hydrologic time series into the time-frequency domain, and have shown promise in estimating timing errors, represented as phase shifts in wavelet spectra, in hydrologic prediction. Here we investigate their application to the National Water Model. Results are presented for several locations in the US with different hydrologic signatures, as well as a synthetic dataset. We point out both advantages and limitations of using wavelets for characterizing event-timing uncertainty and opportunities for improving predictions.
![](images/flag.png)
![](images/blue_ribbon.png)