Date of Award
Master of Science
Geological and Atmospheric Sciences
Kristie J. Franz
William A. Gallus Jr.
The present study examines how skillful probabilistic streamflow forecasts are when using convection-allowing ensemble models’ probabilities of precipitation exceeding specified threshold accumulations as input. Both the High-Resolution Rapid Refresh Ensemble (HRRRE) and High-Resolution Ensemble Forecast version 2.0 (HREF) output were tested. A vital component of this work was the creation of expected rainfall amounts at every grid point for seven different probability of exceedance values. The rainfall amounts for each of the probability of exceedance values were calculated using cubic interpolation from the probabilistic quantitative precipitation forecasts (PQPFs) generated from the HRRRE and the HREF models by use of a Gaussian smoothing technique applied by the HRRRE and HREF developers. The grid point precipitation amounts associated with the probability of exceedance values were then inputted into a hydrologic model for 11 different river basins across the upper Midwest for 109 cases during June, July, August, and September of 2018. It is shown that the process of interpolating PQPFs into the probability of exceedance values and then using them as an input to the hydrologic model produced forecasts that were able to capture the observed changes in the streamflow with a containing ratio of 100%. However, the low probability of exceedance values was associated with discharge values that were extreme, being ~34 times higher than average observed discharge. These high values are likely the result of the approach being too simplistic in that precipitation amounts for a specified exceedance value at every grid point, computed from the PQPFs and were then averaged and input into the Sacramento Soil Moisture Accounting model. Such an approach assumes that all points in the basin would experience rainfall with potentially unusually heavy intensity and longevity. The error in the streamflow forecasts could be counteracted by calibration of the probabilistic derivate precipitation forecasts or by studying the typical distribution of precipitation within convective storms to adjust the rainfall inputs into the hydrologic model.
Goenner, Andrew, "Creating probabilistic streamflow forecasts using HRRRE & HREF probabilistic quantitative precipitation forecasts" (2019). Graduate Theses and Dissertations. 17017.